4.1 Introduction

Language is paradigmatically a human activity, largely consisting of speakers saying things in order to inform, warn, misinform, threat, sell, and so on. Usually, the plans that motivate such utterances include being understood by others and having an effect on their behavior as a result. The use of language belongs squarely in what Frege called the causal realms, the physical and the mental. Language is important because it is a system for doing things. This suggests that the philosophy of action should be a part – a very important part – of the philosophy of language.

To a certain extent it is. Charles Morris, in Signs, Language, and Behavior (Reference Morris1946), divided “semiotics,” the study of signs, into three parts: syntax, semantics, and pragmatics.Footnote ¹ Taking language as a system of signs, pragmatics is the study of how language is used, “how to do things with words,” as Austin says. Austin developed his theory of speech acts in the 1950s, and Searle and others have further criticized and developed it (Reference SearleSearle 1985). Speech act theory categorizes and describes actions in terms of what we do to perform them, and what we do in and by performing them. Perhaps by uttering a sentence I say something about a politician (a locutionary act); in saying that, I insult the politician (an illocutionary act); by insulting the politician I annoy his powerful friends (a perlocutionary act), and so on. All of this depends on syntax and semantics to get started, but it is basically pragmatics.

A bit later, Reference Grice, Cole and MorganH. P. Grice (1975) introduced the concept of implicature, which is something conveyed by an utterance beyond the proposition it expresses, beyond what the speaker says in uttering a sentence in a certain occasion. You ask me if I am going to a talk by X; I say “X is a complete idiot.” I have not said that I am not going but have implicated that I won’t. This, however, is not a matter of logic; I can consistently “cancel” the implicature by saying, “but I promised Y I would introduce X, so I’ll be there.”

Grice intended his influential theory to be a contribution to semantics. He thought that by taking what is implicated to be what is said, we make mistakes about what words and sentences mean. His original targets were claims, “common-place of philosophical logic” (Reference Grice, Cole and MorganGrice 1975: 22),Footnote ² that the truth-conditional analyses logic provides for words like “or,” “and,” “if” do not give their meaning in natural language. But the theory of implicatures, regardless of what Grice intended, is a pragmatic theory, a theory of conversation taken as a “collaborative effort” (p. 26), “a special case or variety of purposive, indeed rational, behavior” (p. 28) where participants observe, or can be expected to observe, rational principles and maxims for (cooperative) action.

By and large, semantics itself was “utterance-free” until about the middle of the last century. Not that philosophers and logicians were unaware of utterances until then. But the working assumption was that semantics, the study of meaning and reference, should focus on what all utterances of an expression or sentence have in common, due to meaning, and not on how they differ, due to the particular facts of the utterance.

In this chapter we first consider how this assumption was challenged when the philosophy of language began to think about natural languages, as opposed to what Frege called “perfect languages.” We discuss how, in the face of these challenges, attempts were made to finesse utterances, in the analyses of names, and indexicals, by creating what one might call “expression-hybrids,” particularly by David Kaplan. While appreciating the brilliance behind these attempts, we express reservations. Then we turn to our own theory, the reflexive-referential theory, which takes utterances as basic to the semantics and pragmatics of natural language.

4.2 Utterances Lost

Discussion of utterances is not common in the philosophy of language developed in the first half of the twentieth century, at least in the analytic tradition inaugurated by Gottlob Frege and Bertrand Russell. The focus was the truth-conditions of sentences, and how they were based on the meanings and references of its parts. It was not the utterances of sentences and their parts that were studied, but the expressions themselves, the meanings they have in virtue of the conventions of the language to which they belong, the references they have in virtue of the world, and the truth-values the sentence of which they are parts have in virtue of all of these things.

In their early works, neither Frege nor Russell were particularly interested in natural languages except as the inspiration for better, more perfect, languages, like Frege’s Begriffschrift, or the language of Whitehead and Russell’s Principia Mathematica were intended to be. They hoped to use these languages to show that arithmetic could be reduced to logic, a doctrine Russell labeled “logicism.”Footnote ³ While their efforts did not fulfill their logicist ambitions, they produced a revolution in logic that had a profound effect on the next generation of logicians and philosophers of language.

Logic was the main interest of Frege and Russell, at least in his early years. Logic is centrally the study of valid inference. This suggests the activity of saying things that are true or are not and of drawing conclusions. But it was not these activities that were the focus, but rather the properties of sentences, considered as abstract objects, independently of their being spoken or written. How can the truth of one sentence, the conclusion, be guaranteed by the truth of others, the premises? Whether the premises are asserted, and if so by whom, when, and where, and whether the asserter is a reliable person with honorable intentions, is not of any interest. The question is whether what the truth of the premises requires of the world guarantees that the requirements for the truth of the conclusion will be met. So, although actual inferences are actions, with all sorts of causes and effects, it is validity as a property determined by structures of an ordered set of sentences that mainly fascinates logicians. And it was topics connected to this issue that dominated the philosophy of language for the first half of the twentieth century, following Frege’s writings on these topics and Russell and Whitehead’s watershed work, Principia Mathematica.

Frege and Russell’s ideas about meaning, reference, truth, and the nature of the propositions gave philosophers of language a lot to think about. It was these ideas, often labeled the “Frege–Russell” theory, that dominated the philosophy of language for the first half of the twentieth century. They are still very much present in contemporary discussions, although their effects are somehow mitigated, or at least combined, with the effects of another “trend” that originated a few decades later, which placed the focus on the study of natural language. This trend, championed by philosophers like Strawson, Austin, and Grice, among others, was not concerned with logic (or not merely) or with the development of any formal language, but rather with understanding the ways in which ordinary languages work and the ways speakers use expressions and sentences to do different things. When the attention of many analytic philosophers turned to natural language, the philosophy of action began to intrude into the citadel of the philosophy of language.

4.3 Reference: Natural Language Intrudes

From the end of World War II on, philosophers of language paid increasing attention to natural language, and to the extent to which the language of logic can mislead us as to how the corresponding expressions are used in natural language. The later Wittgenstein at Cambridge; Urmson, Austin, Grice, Anscombe, and others at Oxford; Peter Geach at Leeds; Max Black, Norman Malcolm, and Keith Donnellan at Cornell; and many others around the world pursued this theme in various ways, seldom in complete agreement with one another. In common to all was the increased attention to action, “how to do things with words,” in Austin’s phrase.

We start by describing the effects of this change of focus to the analysis of singular reference and, specifically, to the classic debates about definite descriptions (which are thought by some to refer, but not by all) and about names and indexicals (which are widely thought to refer to things). We pay special attention to indexicals, which might seem the natural candidates for the philosophy of action to intrude into the philosophy of language.

4.4 The Reflexive-Referential Account

The reflexive-referential account focuses on episodes: utterances, thoughts, perceptions, acts, beliefs, desires, intentions, etc. These things occur in space and time and have causes and effects. We take the relevant utterances to be intentional acts and so call all such episodes “cognitive.” Cognitive episodes have contents. The truth-conditions of a statement or a belief are the paradigm case, but promises and intentions have fulfillment conditions, orders and desires have satisfaction conditions, and so forth. We use “contents” for all such conditions. Contents can be thought of as propositions that encode the conditions of truth for statements, conditions of satisfaction for expressions of hopes and desires, conditions of compliance for orders, and so forth. We take propositions to be abstract objects that are useful for classifying episodes in terms of their conditions of truth, satisfaction, and compliance. We agree with Frege that propositions (“Thoughts”) are not mental or physical episodes or states and so have no causes or effects. We do not, however, accept everything Frege says about propositions.

We cannot, for instance, regard mental states, such as belief, desire, and the like, merely as relations to abstract objects, for beliefs and desires have causes and effects. We think that propositions are abstract objects that have been designed to be useful in classifying events in the causal realm. To be useful, such propositions need to be well-behaved and well-understood abstract objects, usually sets. The language that is used for propositions needs to be correlated with the elements of utterances and thoughts that determine their truth-conditions. Sentences play a major role in identifying these elements in the case of utterances and thoughts that can be expressed with sentences.

On the reflexive-referential account cognitive episodes have multiple consistent layers of content that can be classified with propositions.Footnote ¹¹ Consider the follow example. Robyn and Fred are old friends, but up to now they’ve always lived in different cities. Today, however, Robyn has just moved into Fred’s neighborhood in San Francisco. Suppose this is all happening on June 12, 2021, in a particular neighborhood in San Francisco, say, in the Mission. Robyn knocks on Fred’s door and says to Fred, “Today you and I are neighbors.” Call that utterance u. Then on many accounts, Robyn said the proposition P: that Robyn is Fred’s neighbor on June 12, 2021. On a simple account this proposition is a set consisting of the relation of x being a neighbor of y at t; and the sequence of Robyn, Fred, and June 12, 2021. One might take it to be the set of possible worlds in which Robyn be Fred’s neighbor on that day. Or perhaps the set of worlds w in which Robyn’s counterpart in w be a neighbor of Fred’s counterpart in w on that day, or perhaps its counterpart. Indeed, there are many kinds of abstract objects philosophers have taken to be propositions. We’ll stick with the simple account and assume that the reader gets the general idea.

We agree that the proposition P is a truth-condition of Robyn’s utterance. But we don’t agree that it is the truth-condition. On our account, attributions of truth-conditions are ascribed to an utterance relative to what is taken as given. The truth-condition of an utterance, relative to a set of facts g, is what else has to be the case for the utterance to be true, given g. With regard to Robyn’s utterance, we gave you the facts that determined the references of the parts of his utterance. “I” referred to Robyn, “you” referred to Fred, “today” referred to June 12, 2021, the language was English, so “are neighbors” refers to the relation, between two or more people, of living near to. Given all these facts, what else has to be the case for Robyn’s utterance to be true is

Now suppose we didn’t bother to tell you when this all happened. Do you know conditions of truth of Robyn’s utterance? Of course you do. Robyn’s utterance is true iff

Suppose we also didn’t tell you who answered the door. Then you know that the utterance is true iff

Note that P′ and P″ put conditions on the utterance itself: we call them reflexive or utterance-bound truth-conditions. In contrast, P puts conditions only on the items referred to by the utterance. P gives the referential truth-conditions of the utterance. P could be true if the utterance had never been made. But not P′ and P″.

These three propositions are different and independent. But they are consistent. P′ follows from P″ when we add the fact that the door answerer was Fred. P follows from P′ when we give the date.

Having different propositions, classifying different and incremental truth-conditions, is useful for many things. To begin with, of course, it gives a comprehensive account of what needs to be the case for a particular utterance to be true and of the several layers included in what is said by it. With this explanation at hand, it is quite easy to deal with puzzles in philosophy of language, such as the puzzles mentioned above involving names. Let us look at a couple more examples, to see how it works.

Bob puts a big platter of kale before his young children and says: “You will love this kale.” He knows his children have never seen this ugly vegetable before. But he thinks they will learn the name for this vegetable from his utterance. They will realize that his utterance is true iff there is some stuff Bob is referring to with “this kale” and they will love it. They realize that “kale,” in the singular phrase “this kale,” is most likely a name of the kind of stuff he is pointing at. That is, they know how English works. They do not know this in the way people who have taken a linguistics or a philosophy of language class know it. They may not know what an utterance is, or what referring is, or what a singular term is. But they are attuned to these facts about English and are capable of making inferences from them, without being able to articulate the premises of their reasoning. What explains their inference is an utterance-bound content of Bob’s utterance, given the meaning of the sentence and the fact, conveyed by glance and gesture, that he was referring to the green stuff on the platter. Given all that what else had to be the case for Bob’s utterance to be true? That “kale” is a name for the ugly green stuff. Note that in this case Bob succeeds in teaching his children something, that “kale” names kale, even though what he says is almost certainly false.

John tells Frenchie he is going to San Sebastián in the Basque Country for a week. She looks at the map of the Basque Country in John’s office for “San Sebastián” but cannot find it. He says, “San Sebastián is Donostia.” Frenchie finds “Donostia” on the map, points to it, and says, “That’s where you are going?” and John says “yes.” John has conveyed to Frenchie that “Donostia” and “San Sebastián” are names for the same city. But what he said was simply that San Sebastián was Donostia, a necessary truth, which is the same thing he would have said with “San Sebastián is San Sebastián” and “Donostia is Donostia”, if what is said is always to be identified with referential content. But those utterances would not have been helpful. A philosophical rule of thumb should be that utterances of identity sentences can almost never be understood simply in terms of their referential content. If you insist on doing that, you will be lucky to get much of anything right.

The reflexive-referential theory not only accepts and accommodates the intrusion of the theory of action into the philosophy of language but also takes the theory of action as fundamental to understand how language works. We claim that a theory like this, with utterances at the center and considering them as episodes, that is, as cognitive acts with a variety of contents, is better equipped to deal with the context-sensitivity of language than theories that take (one single) content to be a property of sentences plus contexts.

4.5 Utterance versus Sentence-in-Context

Kaplan discusses utterances to motivate his theory, but he does not bring them into his theory as such; they are replaced by, or perhaps modeled as, pairs of expressions and contexts: “expressions-in-context.” Kaplan’s main interest was developing a logic of demonstratives and indexicals. For this purpose he regarded utterances as an unnecessary complication. For one thing, a pair of context and expression can have a content, even if the speaker of the context does not utter the expression at the time and place of the context in the world under consideration. More importantly, utterances take time; the validity of an argument with one hundred steps might depend on the context being the same for all of them, but we can’t talk or write that fast. So, for logical purposes, utterances can get in the way.

From the point of view of understanding the relations between the contents of states and their causal roles, however, it is very helpful to have episodes that have contents, have causes, and have effects. So, in a theory of language framed in a theory of action, episodes and, in particular, utterances are too important to ignore, in spite of the complications they pose for logic.

For this purpose, we consider the elements of Kaplan’s contexts to simply be properties of utterances, which objects fill the roles of speaker-of, time-of, and location-of. The fact that utterances have speakers and occur in locations at times clearly inspired Kaplan’s concepts of context and character. We promote these inspiring episodes to first-class status. The chief advantage of this view is simply that it accounts for – and makes use of – the fact that utterances have many other properties, in addition to having speakers, locations, and times, that can be relevant to understanding their cognitive significance.

Consider Gricean intentions, which are basically intentions to produce concrete utterances in order to achieve certain effects on the hearer by the recognition of those intentions. The meaning of the sentence and expressions uttered, the content uttered, and the elements of Kaplan’s context all contribute to that goal. But they are not enough. Conversations, linguistic transactions, fundamentally involve a speaker, who intends to do something and achieve certain effects by uttering a sentence or an expression, and a hearer, who perceives an event (or a token of such event) and needs to reason about the intentions behind it. Properties associated with meaning and content are key to guessing intentions, but not only. On many occasions, the hearer has to interpret what the speaker is trying to do by producing an utterance, and not only what she says. The hearer, that is, needs to know if the speaker intends to command something, amuse, impress, mislead, etc.

One way to handle these, while sticking with Kaplan’s approach, is to add more members to the context set, or to introduce additional sets. The latter is more or less the approach of Jon Barwise and John Perry in Situations and Attitudes (Reference Barwise and Perry1983). On their “relational theory of meaning,” the meaning of a sentence is taken to be a relation among various situations connected to an utterance, although the utterance itself is, as in Kaplan’s theory, only modeled and not introduced directly into the theory. The basic relation is between the utterance situation, which determines the speaker, location, and time, and, in lieu of propositions, described situations. But various other situations are added to the range of the relation, to deal with names, descriptions, ambiguity, and other phenomena.

On the reflexive-referential account, however, the treatment of such factors is simpler and more straightforward. They are all properties of the utterance, recognized as necessary to handle various phenomena. The reflexive-referential theory inherits a key insight of Kaplan’s theory – and Perry’s earlier views – but the inclusion of utterances gives it, at least, one important advantage.Footnote ¹² The inherited key insight is the distinction between different ways in which information can be discovered, believed, and asserted.

The advantage is, briefly, that the cognitive significance of an utterance for different hearers can depend on the perceptual and causal relations the hearer has to the utterance, which reference-determining facts they know, and how they think of them. To account for this, we need to bring in additional properties of the utterance and, in particular, causal properties. These are not modeled by Kaplan’s expressions-in-context. This advantage is particularly relevant for us here, since it offers a solution to the cognitive significance problem without having to renounce to the referentialist insights.

4.6 Cognitive Significance

Consider a classical puzzle in semantics. In his important essay “Has Semantics Rested on a Mistake?” (1986), Howard Wettstein pointed out that whatever virtues of what he called the “Kaplan–Perry” account has in explaining cognitive significance of cases involving indexicals, it does not handle Reference FregeFrege’s Begriffsschrift ([1879] 1967) problem, the origin of worries about cognitive significance, which involves proper names. “Hesperus is Hesperus” and “Hesperus is Phosphorus” clearly have different cognitive significance; one learns from the second that the names co-refer, but not from the first. On a directly referential account of proper names, which is more or less what Frege had in the Begriffsschrift, this is hard to account for. On Kaplan’s account, the character of a proper name is a constant function, from any context to the bearer of the name. The two sentences have the same content, a singular proposition to the effect that Venus is Venus. So, whatever the virtues of the Kaplan–Perry thesis for cases involving indexicals, it does not help with proper names.

In response to Wettstein’s essay, Reference PerryPerry (1988) introduced the concept of the proposition created by an utterance, in contrast with the proposition expressed by an utterance, which was basically the distinction between reflexive and referential content.

On the reflexive-referential account, the two utterances have different reflexive truth-conditions. The first is true if and only if there is an object named by “Hesperus” which is self-identical. The second is true if and only if there is such an object, and there is also an object named by “Phosphorus,” and the objects are identical. So, even though the referential truth-conditions of the two utterances are the same, they differ in cognitive significance, in virtue of their different reflexive truth-conditions.Footnote ¹³

Consider now a puzzle involving reference to time. This puzzle involves dates, which are akin to names, and temporal indexicals, such as “today” or “tomorrow.” It is quite common, at least for some of us, not to know the day one lives in. In this case, however, we certainly know when we live; we know that what is happening is happening now or today, even though we do not know what time is “now” or what day is “today.” Also, it is relatively common to be wrong about the date one lives in. One can be in a dateless haze without knowing it; that is, one can be quite certain about the date, but be wrong. In such a case, it is natural to think, in retrospect, that one might have been correct; even that one’s false belief was justified. This provides another advantage for the reflexive-referential theory with respect to Kaplan’s.

Suppose John and Dan are planning to go to the Giants game on August 22, 2021. The day before, John types out a reminder: “The Giants game is tomorrow. Don’t forget.” But he forgets to hit the “send” button. He notices that the message has not been sent just before retiring and hits the button. But he doesn’t notice that it is already after midnight.

Dan, a dateless-hazer, sees the reminder, “The Giants game is tomorrow,” when he wakes up on August 22, and sees that date on the email heading. He knows that the game is on August 22, and reasons, given John’s notorious reliability, “Today must be August 21.” He then immerses himself in linguistic esoterica until late in the evening, when John calls and says, “You missed the game!”

Dan thought, on August 22, that it was August 21. He had good reason for this belief. The game was scheduled for August 22. John said, in an email this morning, that the game was tomorrow. John is very reliable. Therefore, today must be August 21.

Could Dan have been right? It seems not, because for his belief to be true, August 22 would have to be August 21, which is not possible. Surely had Dan said,

1. (4) August 22 is August 21.

we could diagnose some kind of irrationality (or assume he is making some subtle linguistic point). But if he just says, as he did:

1. (5) Today is August 21.

on August 22, this does not seem like the right diagnosis.

There is a way in which way Dan’s utterance could have been true: if it had occurred on August 21 rather than August 22. In other words, if the role of time-of for the utterance of “today” had been filled by some moment occurring on August 21, the utterance would have been true.

On the expression-in-context approach, this does not seem like an option. Since the pair of expression and context is individuated by its members, we would not have the same pair if the time of the context was August 21. So, we have another advantage for the reflexive-referential theory and for granting utterances first-class status in the semantics of tense and indexicals.

There is a possible objection to our strategy, as Richard Vallée has pointed out.Footnote ¹⁴ One theory of events is that they are individuated by the time at which they occur, by where they occur, and by which object and properties involved. If we accept this account of event individuation, the reflexive-referential account is no better off than the expressions-in-context account.

But we reject this account of the individuation of events. In our opinion, it is a plausible account of the individuation of facts, but not of events. Being a fact is a property of whatever one takes to serve as possibilities, be it circumstances, state of affairs, or whatever. We will not here develop an account of events, which we regard as very basic elements of reality. But we think an adequate account must allow for counterfactuals of the form, “if the election had occurred two weeks earlier, Clinton would have won,” or, to follow with our example, “if Dan’s utterance had occurred on August 21, he would not had missed the Giant’s game.” It is a fact that he missed it and that he made his utterance on August 22. Nothing can change that. But the episodes involved, Dan’s belief and Dan’s utterance “Today is August 21” could have been true, had they occurred on August 21.

In his paper “Frege on Demonstratives” (1977), Perry introduced the example of Heimson, who thought he was David Hume. Let’s suppose instead that Heimson thought he was Bob Dylan, which will make it easier to make a case for his rationality. Here is the background story. Heimson falls, hits his head, and has amnesia as a result. He does not know who he is. He carries no identification. He awakes in a hospital where no one has any idea who he is. Heimson decides to figure out who he is. Heimson’s amnesia is of a rather peculiar sort; he retains “third-person” memories about lots of people, he simply does not remember which of them he is. He assumes he is one of the people about whom he knows a great deal. He notices that he knows all of Bob Dylan’s songs by heart, the date of every concert where he performed, and loads of other things. He also knows a lot about a fellow named “Heimson,” but not nearly as much as he knows about Dylan. He decides he is Dylan, and thinks, with some confidence, “I am Bob Dylan.”

His thought cannot be true. Indeed, it seems necessarily false. But it might be rational. And we think it does get at a possibility, even if a rather remote one. The possibility is found at the reflexive level. Call his thought – the event of thinking, the episode, not its content – T. If Bob Dylan had the thought T, rather than Heimson, T would be true. This is the possibility that Heimson’s sifting of the evidence available to him led him to think this was the case.

Reference LewisDavid Lewis (1979), considering Perry’s original example, comes to the opposite conclusion. In “Frege on Demonstratives,” Perry advocated a version of the Kaplan–Perry view, that he called the two-tiered view. In a nutshell, to deal with the attitudes we need to recognize two levels of content for beliefs. What is believed is a proposition, often a singular proposition. How it is believed corresponds to character (Kaplan) or role (Perry).

Lewis had nice things to say about this account but thought that the level corresponding to character or role could serve as what is believed, and the upper tier could be jettisoned. Lewis noted that a character, a function from contexts to propositions, could be regarded as a property: The property an agent has at a time iff the character, applied to that agent and time, yields a true proposition. So, when JP says, “I am sitting,” he “self-ascribes” the property P_sitting:

On Lewis’s view, properties, rather than propositions, are the true “objects” of beliefs. A belief consists of an agent at a time self-ascribing a property.

Lewis’s view, like the Kaplan–Perry approach, does not involve episodes. We have agents, times, the relation of self-ascription, and properties. Reference LewisLewis (1979, footnote 16) regards singular propositions and “de re” beliefs as unnecessary intrusions into the theory of the attitudes based on preoccupation with the analysis of our customs for reporting beliefs.

Lewis’s view incorporates much of the traditional picture that belief is a relation between an agent at a time and a proposition. His innovation is to replace propositions with properties. But we think the idea of objects of beliefs in this sense is a mistake. A belief consists in an agent at a time being in a mental state, an episode. This episode has truth-conditions, which can be characterized by propositions, many different propositions, depending on what is taken as given. But neither propositions, nor characters, nor characters construed as properties, are objects of belief in the sense that belief consists of a relation to them. Propositions, in our sense, are tools we use to characterize and keep track of the truth-conditions of the episode.

Lewis’s account, like the Kaplan–Perry account, does not have episodes, utterances or beliefs, as elements. So, on Lewis’s view, Heimson’s belief, whether he thinks he is Hume or thinks he is Dylan, cannot be true in the strong sense that there is no possible way it could be true. On his view, the belief consists of Heimson, the relevant time, and the attitude of self-ascription to the property of being Hume/being Bob Dylan. Since there is no possibility that Heimson at the time has that property, there is no way the belief can be true. Heimson is irrational in all possible circumstances.

4.7 Conclusion

In this chapter, we have offered a brief and undoubtedly partial account of the ways an account of language as a human activity intruded in the philosophy of language during the past century, focusing, in particular, on the discussion of singular reference. As a way of confronting the need to embrace this intrusion, we presented and defended the reflexive-referential account, and we discussed an example of the beneficial impact of action theory on the philosophy of language.

The conception of language as action inaugurated by Austin, Grice, Strawson, and others not only brought new pragmatic theories of language and novel theoretical tools, such as the distinction between locutionary, illocutionary, and perlocutionary aspects of utterances or the systematic study of nonliteral speech via the notion of implicature, but inspired new philosophical theories, like the reflexive-referential theory or critical pragmatics, to deal with classic puzzles in the philosophy of language like reference and cognitive significance.

5.1 Introduction

The aim of this chapter is to highlight and explain the principal ways in which context is situated within the framework of Systemic Functional Linguistics (SFL) and to explore the main themes and parameters which situate context within an integrated theory of language as a semiotic resource. SFL has viewed context as integral to linguistic theory, rather than an optional extra or a complementary add on. In Butler’s review of three structural-functional theories, he states (Reference Butler2004: 164) that SFL is “the only functional theory to have built in a specific model of social context, through the parameters of field, tenor and mode.” However, not only has context been a core concept in the development of the theory, it also has explanatory power through the SFL concepts of realization, instantiation, and stratification. Within the SFL framework, “context is regarded as explaining the form of the language system as a whole. The clusterings of choices in the language system are hypothesised to derive from clusterings of contextual features” (Reference Berry, Bartlett and O’GradyBerry 2017: 43). As we will show below, the relationship of realization more accurately captures the relationship between context and language, since one is not, strictly speaking, derived from the other in any directional sense. For Halliday, “every act of meaning has a context of situation, an environment within which it is performed and interpreted” (Reference Halliday and Webster1979/2002: 201, emphasis in original).  Not only is language use shaped by context but also context is shaped by language.

Within SFL, context refers to all aspects of a social (communicative) process that have an impact on meaning and shape the outcome of that social process, including the material situational setting, the topic, the modality, and the relationships between the social actors. This view of context is situational and is defined primarily in terms of three parameters (sometimes referred to as variables): the Field of discourse, the Tenor of discourse and the Mode of discourse. Each is described within SFL as follows. Field represents the social context, seen as “the nature of the social process in which the text is embedded – ‘what is going on’” (Reference Halliday and WebsterHalliday 1981/2002: 227); Tenor includes social relationships and roles, where a text is viewed as “the interpersonal relationships among the participants – ‘who are taking part’” (p. 227); and Mode concerns the symbolic organization of language, including “the role assigned to the text, including both medium and rhetorical function – ‘what part the language is playing’” (p. 227). How these parameters relate to semantics and the lexicogrammar within Halliday’s framework will be discussed below.

It is important to note at this point that the field of SFL scholarship is broad and while context will be represented in this chapter following Halliday, his is not the only framework within SFL. For example, Reference FawcettFawcett (1980) developed a different place for context in his model. While the concepts are all very similar in many ways, they do not take up the same place in the framework. For Fawcett, context is a type of knowledge (or information) held by the speaker, and in this sense there are different contexts. For example, Fawcett’s model has a built-in distinction between “context of situation” (recoverable from the observable), “context of co-text” (recoverable from preceding text), and “context of culture” (recoverable from long-term memory). This representation brings Fawcett’s model much closer to the models of context put forward by Philipp Wegener or even Breal (see Reference NerlichNerlich 1990 or Reference Nerlich and ClarkeNerlich and Clarke 1996). Fawcett’s approach to context is perhaps more cognitive and individualistic than Halliday’s more semantic and social one. It is, however, true that the concept of context in Fawcett’s model has not been developed as explicitly as that of Halliday’s. See Reference Fawcett, Fawcett, Halliday, Lamb and MakkaiFawcett (2015) for an overview of his account of context.

While it may be the case that the SFL framework brought context into its theory of language from a very early stage, it was not independent of strong influences both from other linguistic theories and from other disciplines such as anthropology and sociology. This chapter will begin in Section 5.2 with an overview of the historical development of context as a concept within SFL theory. Following this, Section 5.3 examines its place in the SFL theory of linguistics, including how the model accounts for its interaction with language. In Section 5.4 we consider some of the challenges with respect to studies of context: (i) how we can reconcile the place of context within a theory of linguistics and (ii) whether the theory predicts different contexts for different modalities when working with multimodality. Finally, we conclude the chapter with a brief summary and what we feel are key directions for future research in this area.

5.2 Historical Development of Context within SFL

As a functional approach to language, SFL has its theoretical roots very firmly in the European Functionalist tradition (Reference HallidayHalliday 1978), which many argue differs significantly from the American Functionalist tradition including in key assumptions about the nature of language and linguistic evidence (for more on this debate, see Reference Dirven and FriedDirven and Fried 1987 or Reference NewmeyerNewmeyer 2001). While the European tradition is typically traced back to the Prague School,Footnote ¹ clearly the crucial assumptions of European functionalism go back much further and did not arise out of nowhere with the Prague School (Reference Daneš, Dirven and FriedDaneš 1987). What is important for understanding a functional approach to context is the assumptions that these approaches make about language, the nature of linguistic data, and the fundamental role of linguistics as a discipline, since this shapes how theories and models are built and how they are applied in research.

Halliday’s functionalism takes its systematicity from Saussure and Hjelmslev, but with a distinct cybernetics flavor that results in system networks as a means of representation that model both open and closed systems. Reference Halliday and AllenHalliday (1995) is quite clear in relating this to the debate about language evolution:

Functional perspectives on language generally assume that language is a tool used as an instrument to facilitate communication or social interaction. The primary concern is thus on language in use and on better understanding language process and structure, which presupposes modeling both the speaker and the listener.Footnote ² In this view, how language is actually used in communicative contexts becomes crucial. Modeling context, therefore, must be integral to functional approaches to language and all functional models of language must account for context to some extent. However, as Reference Butler and Gonzálvez-GarcíaButler and Gonzálvez-García (2014) argue very few functional models have a detailed model of context, with SFL being a notable exception in this regard.

Perhaps the most significant influence on SFL in terms of context has been Malinowski, who has typically been considered to belong to the macro-sociological or anthropology tradition. However, Malinowski’s account should not be underestimated. In his preface to Malinowski’s book, Reference FrazerFrazer (1922: ix) explains that his method “takes full account of the complexity of human nature. He sees [human nature], so to say in the round, and not in the flat.” While many at that time considered that “pure sociology should confine itself to the description of acts” (Reference FrazerFrazer 1922: ix), it is clear that Malinowski saw the purpose of sociology as being centrally concerned with meaning. In viewing the object of study “in the round,” we see that Malinowski draws from both the micro and the macro in his establishment of the scope of modern anthropology.

The SFL approach has tended to relate context very closely to the grammar and, as will be discussed below, the concept of metafunction (Reference Butt, Wegener, Hasan, Matthiessen and WebsterButt and Wegener 2008), but also through a kind of distributed semantics that allows for a fuzzy boundary between grammar and semantics. Coming to functionalism through his mentor Firth and proto-pragmatists like Wegener, Whittney, and Breal (Reference Nerlich and ClarkeNerlich and Clarke 1996: 248) and directly influenced by Malinowski’s modeling of situated interaction, Halliday’s “context of situation” has always had a distinct place alongside other kinds of linguistic statements. As Reference Halliday and WebsterHalliday (1991/2007) argues, when we arrive at the level of a given context of situation, we are already in the culture – hence, there is no need to proceed to culture. Rather, we have the task of elucidating what we find there in the typical-actual, as Firth referred to it. Halliday’s own practice in this regard appears to be cautious: While he arranges his investigations with respect to field, tenor, and mode, the variables seen as relevant for any given account of context/text stay in close proximity to the register under description.

Firth’s work with Malinowski in London, and Firth’s own fieldwork from Kenya to Afghanistan (Reference ReboriRebori, 2002), confirmed the significance of context and incorporated the notion into a polysystemic, relational theory of choices at many levels. Reference HallidayHalliday (1973) integrated “context of situation” into a linguistic theory that included a separate semantic stratum and a more abstract notion of function – “metafunction.” Context of situation became the interface between language and the socio-material order (or, more correctly, between language and the dimensions of the socio-material order that are of importance to the processes of meaning in a given instance).

Firth’s ideas on context were extended by Reference Ellis, Bazell, Catford, Halliday and RobinsEllis (1966), who provided some clarifications, and Reference Mitchell and MitchellMitchell (1957) who applied it to an actual community. Reference UreUre (1969) brought a typological order to the linguist’s work with semantic varieties, or registers. Work by Reference Hasan and DresslerHasan (1978, Reference Hasan1984, Reference Hasan and Ghadessy1999), in particular, has elaborated and extended the Hallidayan approach by conceptualizing the stratum with explicit motivations for the contrasts within the three major systems of field/tenor/mode, and by developing the systematization of the semantic stratum (crucial to activating the descriptive power of context, through realization and hence, inevitably, in terms of delimitation in specific descriptions).

Halliday’s work represents an important extension to Malinowski’s legacy. Everything in a cultural context may be functional and therefore – in that sense – meaning bearing. Halliday is also, however, demonstrating, through the polysystemic mapping of semantic choice, that function (albeit of an abstract kind) provides the optimum way to understand the internal relations of a language system, not just the externalized tasks to which it is employed. A realizational model is a mapping of patterns of patterns, all of which occur together on different levels of abstraction – from social configuration to modalities of manifestation (e.g., in sound or writing). This evolved “happening together” is a key to the power of languages for extending their potential as systems (or for their speakers to so extend them). Languages are not encodings, but encodings of encodings of encodings (see Reference WegenerWegener 2011 and Reference Butt, Wegener, Hasan, Matthiessen and WebsterButt and Wegener 2008).

While the story of context within SFL developed from the historical roots as just described, it has not remained static. Other theorists, in particular Reference 137Martin, Benson and GreavesMartin (1985, Reference Martin, Cope and Kalantzis1993, Reference Martin, Christie and Martin1997), have extended the stratification above context to genre, and on to ideology, in order to treat variation through Reference GleasonGleason’s (1965) notion of “agnation” (albeit here, an issue of context and of the levels “above” context, not of grammatical variants). There have been numerous other significant contributors to the discussion of context in the SFL paradigm, for example Reference Gregory and CarrollGregory and Carroll (1978) and Reference VentolaVentola (1983), who emphasized the flowing and dynamic nature of context, a notion that we see in Berry’s early work but also recently (Reference Berry2016a), Reference O’Donnell and GhadessyO’Donnell (1999, Reference O’Donnell2021), Reference BartlettBartlett (2016) and Reference MooreMoore (2020). Others have picked up Hasan’s early representations of context as system networks and developed system networks for field, tenor, and mode (Reference ButtButt 1999/2004; Reference Bowcher and GhadessyBowcher 1999; Reference Hasan and GhadessyHasan 1999). Reference BerryBerry (2016b) and Reference WegenerWegener (2011, Reference Wegener, Bowcher and Liang2016) have worked on context for situated interaction and in particular have focused on new divisions for context.

In the next section, we turn our attention to the relationship between context and language, focusing on the way in which context is situated within the SFL theory of linguistics.

5.3 Context and Language

We might want to ask, as does Reference HasanHasan (2014: 47), whether “the study of context deserves the attention and care being demanded hereFootnote ³” since, as she explains, “[i]n modern linguistics until quite recently context had been viewed as an optional extra.” For SFL, the answer is clearly “yes,” since it is one of the few functional theories to have both centralized the concept of context and made of it “an analytical level of description” (Reference Bowcher.Bowcher 2018: 2). The concept of context is theoretically motivated within the SFL framework. Context, as Reference Halliday, Fontaine, Bartlett and O’GradyHalliday (2013: 34) argues, is considered to belong “within the domain of a linguistic theory.” Although central to the theoretical framework, context is extralinguistic for Halliday. Critically, for SFL, context is semiotic and its relation(s) to language is explicit, as will be shown below.

However, we must be clear that context itself, as a semiotic construct, is not “out there,” as in outside the speaker, in any material sense as we might imagine would be the case of the physical surroundings for a speech act. A discussion of semiotic context, as concerns our purposes here as compared to material context, is beyond the scope of what we will consider (but see Reference Bowcher.Bowcher 2018 for a detailed discussion of these two types of context within the SFL framework). While it might be considered outside the language system, context is not independent of language use; it is connected to the language system. Reference Halliday and WebsterHalliday (1991/2007: 281) explains the interaction as follows:

This sense of context being created from text makes clear the inherent relationship between context and language use; language construes context. This perspective on context is summarized very nicely by Reference Berry, Bartlett and O’GradyBerry (2017: 43):

In this sense, we might gloss the relationship between context and language as dialectical in the SFL framework. The definition and place of context within the theory is interpreted somewhat differently by different scholars (see, e.g., Reference MartinMartin 1992; Reference WegenerWegener 2011; Reference MatthiessenMatthiessen 2015; Reference Bartlett, Bartlett and O’GradyBartlett 2017; Reference Tam, Bartlett and O’GradyTam 2017; and various chapters in Reference Fontaine, Bartlett and O’GradyFontaine et al. 2013). However, there is general agreement concerning the significance of four theoretical concepts which contribute to defining, describing, and parametrizing context. These concepts, which we will now explore, are stratification, realization, instantiation, and construal.

Within SFL, stratification is a key pillar within the theory, one of five main dimensions (Reference WegenerWegener 2011). A stratified language system is one which “consists of different coding levels” (Reference TaverniersTaverniers 2011: 1102), where the relationship between the strata is accounted for by realization (see below). It is through the stratified system that language becomes a semiotic system, i.e., one which creates meaning. Stratification is sometimes discussed in terms of abstraction, i.e., higher levels are more abstract than lower ones; for example, context is more abstract than lexicogrammar, although this notion of abstraction is not entirely problem free (for a useful discussion of the concept of abstraction within the theory, see Reference Williams, Russell and IrwinWilliams et al. 2017).

The strata are labeled (from more abstract to more concrete) as follows: Context ⇔ Semantics ⇔ Lexicogrammar ⇔ Expression. This organization is typically represented in SFL literature by a series of embedded circles, each representing a particular stratum.Footnote ⁴ Context is always illustrated as the highest level in order to represent “how the stratified linguistic system is ‘embedded’ in context” (Reference Halliday and MatthiessenHalliday and Matthiessen 2014: 25). We illustrate the stratification of the content plane in Figure 5.1, which is adapted from Reference Halliday, Fontaine, Bartlett and O’GradyHalliday (2013). In the SFL model, semantics is the stratum which serves as an interface between context and lexicogrammar. Reference Halliday, Fontaine, Bartlett and O’GradyHalliday (2013: 34) explains this stratification as follows: “Somewhere between the lexicogrammar and the context – the significant eco-social environment – we recognise an interface stratum, that of semantics, Hjelmslev’s ‘content substance.’” Figure 5.1 represents only a partial view, we have excluded the expression plane (e.g., phonology or graphology) as it does not directly concern our current discussion; however, see Reference TaverniersTaverniers (2011) for more detail about the full representation.

Figure 5.1 A Functional approach to context within a theory of linguistics

(adapted from Reference Halliday, Fontaine, Bartlett and O’GradyHalliday 2013)

As a stratified layer in the theory of linguistics within SFL, Figure 5.1 can be explained as follows: Context is realized as semantics, semantics is realized as lexicogrammar.Footnote ⁵ Equally though, as noted above, text construes context. The relationship between the strata is therefore bidirectional, rather than deterministic. The concepts of realization and construal are discussed in more detail below, as is the dynamic nature of context, but for now we continue with the consequences of stratified context within a theory of linguistics.

As mentioned above, each stratum constitutes a different coding level. Context is organized by the parameters of Field, Tenor, and Mode of discourse, commonly referred to as the variables of register (see Reference Lukin, Moore, Herke, Wegener and WuLukin et al. 2011 and Reference Bowcher, Bartlett and O’GradyBowcher 2017). Semantics is organized metafunctionally, i.e., the broad set of functions that language serves. Reference Taverniers, Thompson, Bowcher, Fontaine and SchönthalTaverniers (2019) offers an in-depth discussion of semantics as an interface stratum, providing considerable insight into how it contributes to language as a meaning-making system. In brief, there are three main metafunctions: Experiential, Interpersonal and Textual. The metafunctional organization of language at the semantic stratum as expressed in the clause is well discussed throughout the SFL literature, perhaps most effectively and most recently by Reference Berry, Thompson, Bowcher, Fontaine and SchönthalBerry (2019). Lexicogrammar represents a “construct of wording” (Reference Halliday and WebsterHalliday 1981/2002: 221); in other words, the form of the content in wordings where lexis and grammar are continuous, inseparable. The organizing principle here is that of rank, with the unit of the clause as the central or core unit. While the concept of rank is largely unproblematic in its broadest sense, i.e. that smaller units make up larger ones and vice versa, the concept of the rank scale as used in SFL is open for some debate (see Reference Fontaine, Schönthal, Thompson, Bowcher, Fontaine and SchönthalFontaine and Schönthal 2019; Reference Berry, Bartlett and O’GradyBerry 2017; and Reference FawcettFawcett 2000). For our purposes here, it suffices to view rank as an organizing principle of the lexicogrammar stratum where the clause, the main grammatical unit, is said to consist of phrase units, phrase units consist of word units, and word units consist of morpheme units.

As mentioned above, the relationship between context and language, as different strata, is that of realization. For Reference Halliday and PutzHalliday (1992: 352) realization is an “interstratal relationship, meanings are realized as wordings, wordings realized as sound (or soundings).” Realization is the principle that accounts for how “patterns on one level both construe and construct patterns on another level” (Reference Cassens and WegenerCassens and Wegener 2008: 208). While realization may, at first glance, seem somewhat similar to the relations described above for the rank scale, it is not, as Reference Wegener, Cassens and ButtWegener et al. (2008) explain, a “consists of” type of relation. Attempts have been made within the SFL literature to account for the relation between context and metafunction by means of a so-called context-metafunction hook-up hypothesis, or CMHH (see Reference Hasan, Fries and GregoryHasan 1995). However, as Reference Berry, Bartlett and O’GradyBerry (2017: 47) points out, despite a lack of evaluation, this hook-up hypothesis is generally accepted as fact. The main problem according to Berry (p. 47) is as follows: “if contextual features are established via semantic features and semantic features are established via contextual features, the reasoning is circular and the hypothesis becomes untestable.” While there is no space to consider these issues here, Reference Berry, Bartlett and O’GradyBerry (2017) lists several works for the interested reader, including Reference Thompson and GhadessyThompson (1999), Reference Clarke, O’Grady, Bartlett and FontaineClarke (2013), and Reference BerryBerry (2016b).

In Figure 5.2, which situates context and language in the SFL framework, we can see that realization is the principle accounting for the relationship between context and language along the stratification dimension. This means that context is “realised in / construed by” the stratum below (Reference Halliday and WebsterHalliday 1991/2007: 275), which, in the stratified model given in Figure 5.1, is semantics. The same can then be said for semantics, i.e., that semantics is realized in/construed by lexicogrammar, and so on. Therefore, an inherent aspect of realization is the SFL concept of construal.

Figure 5.2 Language and context, system and instance

(adapted from Reference Halliday and WebsterHalliday 1991/2007: 275)

Construal,Footnote ⁶ as a theoretical term in SFL, is what gives meaning to the instances of language. For example, an imperative clause such as eat your vegetables is said to construe a command. This command, in turn, would be said to construe a feature of the context parameter of Tenor, for example construing an unequal power relationship (e.g., social hierarchy). Reference Lukin, Moore, Herke, Wegener and WuLukin et al. 2011 provide a detailed discussion of construal as it relates to realization between the strata of context and semantics. Construal is, then, the process of meaning-making; we might think of it as a kind of inference, or interpretation. According to Reference TaverniersTaverniers (2011: 1122), construal refers to “the relationship between language and extralinguistic reality” and is therefore an essential component of inter-stratal realization. Reference Hasan, Mahboob and KnightHasan (2010: 12) explains this relationship as follows: “In the reception of the utterance, realisation is construal of the relevant choice at the higher level: thus, in decoding an utterance, the choice in wording construes meaning, the choice in meaning construes context.” Here we can see that realization and construal are nearly equated in terms of, in the one sense, relating the wording to the semantics, but we also find simultaneously that the wording construes meaning, i.e., it is meaning-making. For further discussion of construal and how it relates to other concepts touched upon briefly here, see Reference Berry, Bartlett and O’GradyBerry (2017).

Having considered realization and construal in relation to the stratification of context, we also note in Figure 5.2 a relation between context as culture and context as situation, which are differentiated by the principle of instantiation. In contrast to the inter-stratal relationship of realization, instantiation is an intra-stratal relationship of the move between the system and the instance (Reference Halliday and WebsterHalliday 1994/Reference Halliday and Webster2002: 352). Reference TaverniersTaverniers (2011: 1120) explains that this principle “concerns the relationship between a schema and an instance (a particular usage/an instantiation) of this schema.” Reference Berry, Bartlett and O’GradyBerry (2017: 53) describes instantiation as “a relationship between the theory and the data.” In this view, context of culture as a theoretical instantiation holds a significant place in the SFL model of context in that, as Reference MatthiessenMatthiessen (2015: 34) explains, the two perspectives on context are differentiated by instantiation: “context of culture is located at the potential pole, context of situation at the instance pole.”

For Reference Halliday and WebsterHalliday (1991/2007: 282) “the culture is construed by systems of language choice; the situation is construed by patterns of language use.” However, he warns that in talking about these two aspects of context, we must not slip into thinking that these are two different things. The representation in Figure 5.2 of the relationship between system and instance is meant to show one and the same object of study from two different perspectives. The system perspective offers one of meaning potential, in other words “the potential that lies behind all the instances” (Reference Halliday and WebsterHalliday 2002: 10). Space does not permit a detailed exploration of culture here but see Reference LukinLukin (2017) for a more detailed discussion of context of culture together with its relation to ideology. For our purposes here, we can think of instantiation as a type of abstraction, where instance (here context of situation) is more concrete, and potential (here context of culture) is more abstract. As Reference Williams, Russell and IrwinWilliams et al. (2017: 14) explain, “The notion of potentials is fundamental in systemic theorising, because it allows the linguistic model to apply not just to previously encountered linguistic utterances, but also to additionally try to explain how speakers produce and interpret novel instances of language.”

The view we now have of context in the SFL framework is that context is realized in language, and also construed by it. As Reference O’Donnell, Tucker, Huang, Fontaine and McDonaldO’Donnell (2020: 73) explains, “Context of situation is thus defined in relation to language, rather than being defined only in relation to the social system itself.” As a stratified layer in the framework, the relationship between context and language is mediated by semantics.

Reference Halliday and WebsterHalliday’s (1991/2007: 282) claim that “text and situation come into being together” can be interpreted as text and context co-constructing. Context is not deterministic, nor is it static or stable. This perspective on context is captured very well by Reference O’DonnellO’Donnell (2021) as follows:

We have focused on the strong connection that the SFL notion of context makes with language. The theory views the language system as stratified. In this sense, we find interacting layers or strata which account for how meanings are construed in contexts of language use. Context, while viewed as extralinguistic, is nevertheless implicated in the theoretical linguistic framework by the realization principle which provides the account of how context relates to lexicogrammar and at this interface is where we find the stratum of semantics.

In the next section, we examine this framework for context in more detail by considering two aspects which pose theoretical challenges. In doing so, we also provide a sketch of important directions for future research and development.

5.4 Two Challenges in Working with Context

While in this section we give our attention specifically to the perspective of SFL, the two areas we discuss are in some ways common to all functional approaches and applications, at least to some extent. Of particular interest for us is (i) where context sits in relation to language and (ii) what the relationship is between context and different modes of meaning-making behavior.

5.5 Concluding Remarks and Future Directions

SFL has always had a transdisciplinary and applied focus, something which has shaped and will continue to shape the theory. As we see models of context applied across more varied domains and engaging with different disciplines, we see the theory taking new paths in new directions. For example, applications in domains such as health and medicine (Reference Moore, Thompson, Bowcher, Fontaine and SchönthalMoore 2019), education and learning (Reference Byrnes, Thompson, Bowcher, Fontaine and SchonthalByrnes 2019; Reference Mickan, Thompson, Bowcher, Fontaine and SchonthalMickan 2019), and computing and artificial intelligence (Reference Bateman, McDonald, Hiippala, Couto-Vale, Costetchi, G. Thompson, Bowcher, Fontaine and SchonthalBateman et al. 2019; Reference Wegener, Cassens, Kaltenbacher and StöcklWegener and Cassens 2019; Reference Zappavigna, Thompson, Bowcher, Fontaine and SchonthalZappavigna 2019) have forged new pathways for the development of context modeling in SFL.

It is fair to say that research drawing on the SFL framework has generally focused to a large extent on the social aspect of language and context. More recently we have seen a return to some consideration of the individual. This more cognitive aspect was always available (consider, for example, the work of Philipp Wegener and even Firth and Ellis), but it was much less prominent and underdeveloped to the extent that one might argue it was completely ignored. Recent work from Martin on individuation (Reference Martin, Thompson, Bowcher, Fontaine and SchonthalMartin 2019) and Wegener on a person-centric approach to context for human-centered AI (Reference Wegener, Bowcher and LiangWegener 2016; Reference Wegener, Cassens, Kaltenbacher and StöcklCassens and Wegener 2019) has renewed this focus on cognition and the person, in different ways and for different motivations.

It may be perceived by some that after what may seem like an extended period of focus on the social aspects of context, the cognitive aspects have still not made an impact on the development of theory. The question of having “an overall cognitive theory of context as a type of mental model” (Reference van Dijk, Wodak and Chiltonvan Dijk 2005; see also Reference van Dijkvan Dijk 2006) is a worthwhile one. Far from treating mental aspects as alien, SFL provides a venue for examining the features, which combine to produce the structural nature of mental aspects such as purpose. No theory has all the answers to all the questions, but if, as Reference Butler and Gonzálvez-GarcíaButler and Gonzálvez-García (2014) suggest, there is fertile ground to explore, it would best be done collaboratively, since each functional theory brings its experience and developments from their own extended periods of focus whether that concerns cognitive processes, typological comparisons, or text in context.

If we accept that context cuts across all strata of language, and as such aids the understanding of meaning at all levels, then we can define it simply as everything that is around the text. Context can, and should, be used to understand the linguistic process; the design of research questions, the collection, storage, and analysis of data, and the interpretation and representation of results. In this sense it is both theory and method. As textual analysis, multimodal or otherwise, is carried out, it is context that makes any findings meaningful. Context does not need to be mysterious or idealized, but by the same principle, neither is it helpful to reduce or oversimplify. Perhaps what is most needed in future work is a clear setting out of a process, one that supports the shift from theory to practice, while enabling and supporting collaboration and transdisciplinary work on context. There are two key areas for such a development.

Firstly, there is a need for greater transparency around context. While there is a relatively clear process for carrying out grammatical analysis or morphological analysis, and even to a certain extent semantic analysis, analysis at the level of context has no unified or clear procedure. There are no established units, no guidance as to how one might set about analyzing a text in context with the result that we usually end up with a very discursive description of context.

Secondly, without a clear procedure it is almost impossible to combine analysis from different projects for comparison or for complementation. At the moment, if one wishes to compare two contexts, it is necessary for the analysis to have been done using the same approach and preferably by the same researcher. Ideally it should be possible to compare across research projects no matter who the researcher and no matter what approach was taken. If our comparison is made on a functional basis, this should be achievable. Just as Reference Halliday and MatthiessenHalliday and Matthiessen (2004) compare the traditional grammatical terms with the functional ones, allowing comparison of analysis, so it should be possible to compare different contextual analyses if we know the functional motivation of the distinctions. This process is an exercise in ontology matching, which, as Reference WegenerWegener (2011) explains, is useful since there are definite limits to what can be considered context. Different theoretical approaches and different models within theories differ not so much on what they consider context to be, but on how they organize their description. It is possible to see how these different approaches have arranged information in different ways, foregrounding some aspects and hiding others but in many respects covering the same ground. Reference GrimshawGrimshaw’s (1994) complementary studies of professional discourse show this very clearly.

Ultimately, the biggest challenges for the future of context in SFL seem to be modeling and representational challenges. With the exception of some small difficulties, the theory itself provides a useful space for context, and applications of SFL across varied domains make use of this, but it is still a rather case-specific application. Some solutions will undoubtedly be technical, particularly with respect to corpus work, particularly large-scale multimodal and multiparticipant corpora. But much of the future work lies with modeling context to make that connection between theory and practice. New tools change theory and practice, just as tools change us as humans and individuals.

We would like to end this chapter on context with the following words from Reference Halliday and WebsterHalliday (2003: 3):

6.1 Introduction

There are two basic perspectives to be taken on a clause. It can be seen principally as a hierarchical structure or as a succession of units. Hierarchical syntactic analysis is most saliently exemplified in contemporary linguistics by phrase-structure trees. These serve to bring out language users’ tacit knowledge of relationships among constituents of the clause. They show which elements are closely related and which are more distantly connected. Hierarchical analysis can take the form either of constituency diagrams that display relations of sisterhood between co-constituents of a hierarchically higher constituent or of dependency diagrams that show the hierarchy of head-modifier relations between elements. Such hierarchical representations are particularly prominent in approaches to syntax that focus on linguistic competence: Their aim is to lay bare language users’ underlying knowledge of the syntactic potential of their language. In some such approaches, the syntactic analysis is limited to indicating relationships of constituency or dependency, with the left-to-right (or earlier-to-later) order of elements being regarded as a matter of phonology.

By contrast, an emphasis on the succession of syntactic units is more typical of grammatical approaches that foreground performance and language processing. Here the focus is on representing the timecourse of the utterance, the gradual, incremental build-up of the clause as one component unit follows the other in real time. While an emphasis on hierarchy tends to be associated with grammars that abstract away from contexts of use and accentuate the grammaticality of sentences, the incrementalist approach generally views language use as situated action and stresses the achievement of situation-appropriateness. This involves a consideration of how an utterance reflects a language user’s communicative goals and needs and the context in which they are conversing with their interlocutor(s). Whereas the perspective that privileges the attribution of a hierarchical structure understandably focuses chiefly on complete clauses and relies heavily on intuitive judgments of well-formedness, the alternative incrementalist view is more open to including incomplete or fragmentary clauses in the purview of syntax and to drawing its data from transcriptions of verbal interaction.

The present chapter aims to bring out how such incrementalist approaches to grammar have developed in tandem with psycholinguistic modeling of language production and comprehension. An important emphasis of the chapter will be on how aspects of the multisensory situational context are rapidly integrated into language users’ processing strategies. Against this background, we will see that research has been moving in the direction of an integration of the two perspectives distinguished in the opening paragraphs. In terms of the structure of the chapter, Section 6.2 will sketch how work on contextually embedded language processing has had implications for the well-established hierarchical analysis of syntactic units, while Section 6.3 will consider various approaches that privilege incremental processing. Section 6.4 will turn to findings relating to the emergence of hierarchy from sequence and thus the integration of the two perspectives, and Section 6.5 will close the chapter by considering the matter from the viewpoint of dialogue.

6.2 Hierarchical Structure Defended, Curbed, and Challenged

The strongest measure of hierarchization in a constituent structure tree is achieved by uniformly imposing binary branching, i.e., limiting the number of co-constituents (“sisters”) to two. This is notably characteristic of Chomsky’s Minimalist Program (Reference BoeckxBoeckx 2006) within generative grammar, as a result of the generalized application of the operation Merge, which combines exactly two elements, a and b, to form a set b {a, b}, where a and b are immediate constituents of a more embracing unit b. As Reference Kornmesser, Kertész, Moravcsik and RákosiKornmesser (2019: 505; italics in original) points out, “postulating merge to take at least and at most two syntactic objects and hence to produce binary branching structures is considered to satisfy the requirement of economy.”Footnote ¹ Several assumptions of generative grammar, minimalist or not, notably the presence of various “empty” elements in the tree structure and the analysis of certain morphological phenomena as “syntactic” and as binary-branching phrases, as well as the absorption of semantic phenomena such as Tense and pragmatic phenomena such as Topic, Focus, Contrast, and even (illocutionary) Force into syntax, again in the form of branching phrases, all contribute to a very high degree of hierarchization. Extreme hierarchization is particularly evident in so-called cartographic approaches to generative syntax (e.g., Reference CinqueCinque 1999; Reference RizziRizzi 2004), which entail abundantly branching analyses, with over thirty “projections” just for adverbial modifiers.Footnote ²

Ray Jackendoff, once a major figure in generative syntax (Reference Jackendoff1977), has come to reject such “syntactocentric” approaches, proposing (Reference Culicover and JackendoffCulicover and Jackendoff 2005; Reference Jackendoff and AudringJackendoff and Audring 2020) alternative models (Simpler Syntax and Relational Morphology respectively), in which constituent structure is retained but no longer restricted to binary branching and is freed from non-syntactic notions. In this approach, syntax is one of three modules of a Parallel Architecture (phonological structures; syntactic structures; semantic/conceptual structures) that coexist in the grammar. For comparable but distinct architectures of this kind, see Functional Discourse Grammar (which has four modules, one of them devoted to pragmatics; Reference 158Hengeveld and MackenzieHengeveld and Mackenzie 2008) and the Autonomous Modular approach of Reference SadockSadock (2012; with six modules). Reference Culicover and JackendoffCulicover and Jackendoff (2005) defend what they call their “relatively flat” (p. 108), multiple-branching structures, and as they point out, these structures are analogous to what has been proposed in other offshoots of the generative paradigm such as Lexical Functional Grammar (LFG; Reference BresnanBresnan 2001),Footnote ³ Head-Driven Phrase Structure Grammar (HPSG; Reference Pollard and SagPollard and Sag 1994)Footnote ⁴ as well as the more practical grammar of English edited by Reference Huddleston and PullumHuddleston and Pullum (2002); to quote an observation by Reference Culicover and JackendoffCulicover and Jackendoff (2005: 107) that will be significant for later sections of this chapter, an emphasis on sequence rather than hierarchy reflects “the view of syntax generally adopted in psycholinguistic research.”

This line of thought is followed up to an extreme degree by Reference Jackendoff, Wittenberg, Newmeyer and PrestonJackendoff and Wittenberg (2014), who argue that there are certain forms of linguistic organization that dispense with hierarchical structure altogether. They defend the view that in pidgins, late L2 acquisition, “home signs” (i.e., spontaneously arising gestural languages), agrammatic aphasia, and even certain fully functional languages such as Riau Malay, the sequence of syntactic units is purely linear, with no hierarchy being detectable. This is a situation which they model by having semantic/conceptual structure map directly onto phonological structure without any syntactic structure coming into play (i.e. an architecture with only two modules). The ultimate position along this trajectory of thought – the claim that syntax is not needed for any language – is defended by Reference NordströmNordström (2017), who argues that Merge is not implementable and proposes what she calls a “flat structure” (p. 697), which, again significantly, she qualifies as being “much more in line with language production and comprehension” (p. 702). Reference SteedmanSteedman (2022: 22) observes in addition that “[m]any languages have freer word order than English, and do not support any clear notion of surface constituency. Even for English, there is no clear consensus on whether the surface structure of the ditransitive VP or the subject auxiliary-inverted question is flat and ternary-branching, left-branching, or right-branching.” Reference Everett, MacWhinney and O’GradyEverett (2015: 361) has written in this regard that Merge “is potentially falsified by any exocentric or non-binary- (ternary-, quaternary-, etc.) branching structure, e.g. a structure with flat syntax.”

Although these various proposals to restrain or even abandon hierarchical structure have largely come from critics who share generative grammar’s orientation to competence, it was curiously the appearance of a short paper by Reference Frank, Bod and ChristiansenFrank et al. (2012) that caused the greatest consternation in generative circles. The introduction to that article makes it clear that the authors do not doubt the importance of hierarchical structure for the description of languages and are merely questioning its relevance for the study of use, which is a matter of performance, i.e., the second approach distinguished here. Nevertheless, the paper’s impact was considerable, and the leading generative grammarian, David Pesetzky, confronted it directly in his plenary lecture to the 2013 Linguistic Society of America’s annual meeting, speaking of Frank et al.’s paper as a “nightmare” and citing Chernyshevsky’s (and Lenin’s) not entirely rhetorical question Что дѣлать? (What Is to Be Done?) in response. It is beyond doubt that grammarians who see themselves as mentalists, as unveiling cognitive structures through their study of language, are sensitive to findings from psycho- and neurolinguistics, and these findings are that binary-branching hierarchical constituent structure is not a given: “[r]ather, considerations of simplicity and evolutionary continuity force us to take sequential structure as fundamental to human language processing. Indeed, this position gains support from a growing body of empirical evidence from cognitive neuroscience, psycholinguistics and computational linguistics” (Reference Frank, Bod and ChristiansenFrank et al. 2012: 4530). As we shall see in the following section, there are scholars out there, a notable example being Reference Christiansen and ChaterChristiansen and Chater (2016), who are willing and able to shake off what they call “Chomsky’s hidden legacy” (p. 6) of focusing on linguistic competence and to explore the implications of not separating the study of grammar from the study of language processing in context.

6.3 Elements of Incremental Syntax

Although, as we have seen, mainstream currents in syntax have emphasized constituency (or dependency) and hierarchical structure, there has always been a strong undercurrent of work that has privileged incrementality. This has tended to go hand in hand with the assumption that each increment leads to a reduction in information entropy, understood as the information that the recipient lacks at any point in the construction of a discourse world. Just as, in Groenendijk and Stokhof’s words (Reference Groenendijk, Stokhof and Brown2006: 29), “the interpretation of a discourse – a sequence of sentences – takes the form of an incremental construction of a discourse representation structure,” so within a sentence, too, the production and comprehension of the various components have been seen as similarly incremental. This notion has found expression in a range of different theoretical initiatives, for example in the Grammar of Speech of Reference BrazilBrazil (1995), which reflects an actional view of syntax, seeing the syntactic unit (chiefly the clause) as “something that begins, continues, and ends in time: it happens” (p. 11). In this view, syntactic structures are viewed as the result of the gradual, piecemeal construction of utterances in time. The initial state (i.e., the first increment) sets up a framework of expectation that ultimately leads to the target state, i.e., the final increment. Brazil’s work was continued and developed by Reference Sinclair and MauranenSinclair and Mauranen (2006) in their Linear Unit Grammar, which, as its name suggests, focuses on syntagmatic relations within clauses but goes further in examining how certain successions of units are linked together to form “chunks.” They stress (pp. 130, 134) that chunking is a “pre-theoretical term” and not an exact science in the way that constituent structure analysis seeks to be exact. In fact, they see the very flexibility of their conceptual framework as advantageous in permitting cross-speaker variation within an overall speech community. This emphasis on the practical benefits of approximation will return in various psycholinguistic approaches that encompass the notion of “good enough” processing.

This perspective on syntax engenders a particular view of minimal or holophrastic utterances, those that consist of one chunk only. Whereas these have been seen from the viewpoint of hierarchical syntax as reductions of full clauses, the incrementalist position views them as complete within the context in which they occur. For example, in one-chunk answers to questions, the question itself provides a large amount of the context in which the brief answer will be understood. Similarly, holophrastic greetings will typically elicit an equally holophrastic response. The ubiquitous occurrence of minimal utterances in conversation follows, given a sufficiently rich context, from the Gricean Maxim of Quantity, which enjoins the language user to invest a linguistic expression with no more (and obviously no less) information than the interlocutor requires for satisfactory interpretation. Reference Levinson, Papi and VerschuerenLevinson (1987: 68), too, proposes a Maxim of Minimization “Say as little as necessary,” that is, “produce the minimal linguistic clues to achieve your communicational ends.”

Many utterances in adult language are of course not minimal in this sense, and the logic of the incrementalist approach is that these are sophistications of the holophrase rather than the latter being an elliptical version of the former. As will be immediately clear, this direction of argument recapitulates the familiar series of stages in child language acquisition, where the child initially produces single-word utterances (“the holophrastic stage”) and then progresses gradually to more complex utterances. Reference Greenfield, Reilly, Leaper, Baker and BarrettGreenfield et al. (1985) apply a Principle of Informativeness to the child’s holophrastic stage, which states that the child will verbalize new rather than given information. At the later two-word stage, the child expresses information that, at the preceding stage, could only be inferred by the addressee from gestural or intonational means and/or by drawing on the context. The underlying assumption is that we communicate in order to draw attention to variability in the environment, that is, what is novel, changing, or alternative. For example, if a mother rolls a ball to her infant, the latter will utter ball rather than Mummy, because it is the ball that is in movement and undergoing transfer, while the mother is relatively static and her agency is presupposed rather than asserted. This insight applies equally well to the adult use of holophrases: for a minimal utterance to function as complete within its context, it must contain the one linguistic clue from which the hearer can read off or infer the crucial information that the speaker wishes to impart. For conversation, this notion is present in Reference ChafeChafe’s (1994: 119, 159) vision of each unit making a “single point” (in both English and the Indigenous language Seneca) and, for writing, in Reference GivónGivón’s (1995: 358) “one-chunk-per-clause” constraint, which states – again using the above-mentioned term of art “chunk” – that “[c]lauses in natural text tend to have only one chunk (usually a word) of new information per clause.”

It is worth stressing at this point that it is not just holophrastic utterances that rely on context for full comprehensibility. Full clauses, too, can only be understood in their entirety if they are submitted to inferential interpretive processes: The hearer does not merely incrementally decode (or “parse”) the incoming utterance but also interprets it incrementally in its context in line with his understanding of the ongoing and dynamic discourse situation. This suggests a fundamental continuity between minimal and maximal utterances but also raises the question of how more sophisticated utterances differ from holophrases in their communicative content. The connection lies in the distinction between assertion and presupposition: While a (declarative) holophrase offers only assertion, leaving the presupposed information to be inferred from the context, the fuller utterance contains a combination of asserted and presupposed information. The traditional term “assertion” will, however, not be retained here because it is so strongly associated with declarative illocutions (“statements”). A preferable term is the information-structural notion “focus,” which is neutral with respect to illocutionary distinctions (Reference 158Hengeveld and MackenzieHengeveld and Mackenzie 2008: 89 ff.). In a wh-interrogative, for example, be it holophrastic or not, it is the wh-word that functions as its focus, in the sense of identifying the particular piece of information to be elicited from the addressee. Consider (1), cf. Reference Mackenzie, Aarts, Bowie and PopovaMackenzie (2020: 193):

1. (1)  [Who]_focus [were you talking to just now]_{presupposition}?

Here the speaker fills out the utterance by reiterating the interactants’ shared knowledge that the hearer was talking to someone in the immediate past and thereby creates a context for the interaction. SheFootnote ⁵ could also have invoked the context with greater reliance on inference, as in (2), with the anaphoric pronoun that:

1. (2)  [Who]_focus [was that]_{presupposition}?

Or she could even have formulated it holophrastically, perhaps with a supportive gesture, gaze, or head movement, for example, if the question is whispered surreptitiously to the hearer:

1. (3)  [Who]_focus?

He similarly has various answering options:

1. 1. a. [I was talking to]_{presupposition} [Bill]_focus [just now]_{presupposition}

   2. b. [I was talking to]_{presupposition} [Bill]_focus

   3. c. [To]_{presupposition} [Bill]_focus

   4. d. [Bill]_focus

Since the information about the conversation with Bill not only is shared but has also been explicitly mentioned in the question, the holophrastic answer in (4d) or the near-holophrastic one in (4c) is sufficient. The presupposed information can be repeated in part as in (4b) or in whole as in (4a). But to do this, especially in full, is to opt for the marked option in conversation, which may spark off implicatures (i.e., indirectly communicated meanings), for example that the speaker of (4a) is irritated by the question.

It is useful at this stage to bring in a crucial notion from Functional Discourse Grammar (FDG; Reference 158Hengeveld and MackenzieHengeveld and Mackenzie 2008), that of the discourse act. A discourse act is a unit of linguistic action that may take the outward form of a full clause, a sequence of phrases that is less than a clause, or a holophrase. Each discourse act is made up of a series of subacts, which together constitute the communicated content of the discourse act. One of these subacts bears the focus of the communicated content (cf. Reference ChafeChafe’s 1994 “single point” or Reference GivónGivón’s 1995 “one chunk” referred to above), while the other subacts, if any, are backgrounded. FDG’s more action-oriented approach allows us to move away from counting only single words as holophrases: They are not identified in morphosyntax but rather at the level of the pragmatic activities of reference and ascription (Reference SearleSearle 1969). Just as children at the holophrastic stage may use certain word combinations as unanalyzed wholes (allgone, allfalldown), so formulaic or prefabricated units (of different degrees of rigidity) also play an important role in adult usage, being interwoven with more compositional elements. Reference WrayWray (2002: 8) points out that “[o]nly with the new generation of grammatical theories, based on performance rather than competence …, has the idea of holistically managed chunks of language been slowly reinstated, and its implications recognized.”

FDG adopts the following compromise with regard to the hierarchy‒sequence debate (Reference Mackenzie, Carnie, Siddiqi and SatoMackenzie 2014a): Syntax is seen as primarily linear, while pragmatic and semantic organization are predominantly hierarchical, with the hierarchies in question representing scope relations (e.g., subacts scoping under discourse acts, or states-of-affairs scoping under propositional contents). Now, when it comes to the syntactic description of incrementally composed utterances, FDG recognizes various salient (so-called “absolute”) positions in the linear order of the various morphosyntactic units, first and foremost among which is the initial position (P^I). Morphosyntactic units that correspond to the upper reaches of the pragmatic and semantic scopal hierarchies that describe the formulation of the discourse act have prior access to those absolute positions. For example, discourse markers, which are analyzed as modifiers of the discourse act, tend to be placed in initial position by virtue of their high position in the pragmatic hierarchy, which represents wide scope.

Returning now to holophrases (or one-chunk utterances), these correspond to discourse acts whose communicated content contains only one subact, necessarily (or “trivially”) marked as focus. They thus rightfully claim the absolute initial position, even in the absence of further material. Thus, all the following examples will be assigned to the initial position of the syntactic unit:

1. 1. a. Brilliant.

   2. b. Goal!

   3. c. Arthur!

   4. d. Ball. (as in holophrastic-stage child speech)

   5. e. Allgone. (as in holophrastic-stage child speech)

Things become more interesting where the communicated content contains more subacts, as in the following examples:

1. 1. a. That was brilliant.

   2. b. They scored a goal!

   3. c. Hey Arthur!

   4. d. Mummy ball. (as in two-word-stage child speech)

   5. e. Banana allgone. (as in two-word-stage child speech)

What we see in (6), as compared to (5), is that units expressing other (non-focus) subacts have usurped P^I, pushing the focus unit to a later position in the sequence. Here the “minimization” advantage of the smallest possible expression in (5) has been abandoned in favor of providing sufficient contextual grounding for the hearer: in (6a), for example, the initial that, possibly accompanied by a gesture, provides the hearer with some indication of what the speaker finds to be brilliant. However, there are advantages for the speaker, too, in formulating a discourse act non-holophrastically. As Reference Wagner, Féry and IshiharaWagner (2016) points out, “[s]tudies from the production planning literature [in psycholinguistics, JLM] have found that one crucial factor in such word order choices is that speakers tend to realize constituents encoding more active and accessible referents earlier in the sentence.” In other words, speakers can in practice gain precious milliseconds from not encoding the focus immediately, but anchoring it in knowledge that is already “active” (i.e., energetically present in the common ground) and accessible (i.e., relatively easily retrieved, for both parties). In FDG terms, such subacts are known as topics, defined technically as “signalling how the Communicated Content relates to the gradually constructed record in the Contextual Component” (Reference 158Hengeveld and MackenzieHengeveld and Mackenzie 2008: 92), where the Contextual Component is an extra-grammatical module that implements the notion of a shared “common ground” of contextual assumptions.

Since topic, like focus, is assigned to a subact, the two are found at the same hierarchical level of pragmatic organization, and therefore if syntactic order reflects hierarchical status, languages may be expected to differ in which “wins out” in terms of access to the initial position P^I. In English, generally, it is the topic unit – which is often also the clausal subject – that goes to initial position, pushing the focus unit further to the end. By contrast, in Nootka (Reference Fortescue, L. Mackenzie and Gómez GonzálezFortescue 2004: 155), the element considered by the speaker to be the most “newsworthy” (a notion that correlates with focus status) remains in initial position, in keeping with a “strategy … found in many North American languages with pragmatically controlled word order.” In English, too, of course, there are constructions in which the focus retains its initial positioning, most notably in wh-interrogatives, where the wh-word – cf. (1)–(3) above – is the focus of the question. This also applies in what Reference HannayHannay (1991) calls the “reaction mode” of declarative discourse acts, as in (7):

1. (7)  Brilliant that was!

where accessible information (that) is mentioned but is denied the status of topic, since (7) is not presented as being about “that.” Since the only pragmatic function present in (7) is the focus, it has no rival for initial position and that is where it appears. The phonological phrase that was in (7) can feel almost like an afterthought to a holophrase and is accordingly pronounced on a low, flat contour.

The approach being sketched here applies without further ado to declaratives, be they exclamative or not. However, with other illocutions (interrogative, imperative), in English it is the marker of illocutionary status that appears in P^I, reflecting the fact that in pragmatic organization these are hierarchically higher than the subacts that carry the functions topic or focus. The result is that any topic unit is pushed to a later position and the focus unit even later:

1. 1. a. Does (P^I) that (Top > P^I+1) matter (Foc > P^I+2) ?[Interrogative]

   2. b. Give (P^I) me (P^I+1) a pen (Foc > P^I+2) ! [Imperative]

The effect of postponing the focus unit can lead to an analytical decision to associate focus status (at least in certain configurations) with clause-final position (P^F), where the language user’s strategy is geared to exploiting the very prominence of that final position (a strategy recognized in the literature as “end focus”).

6.4 Incrementality and Prediction in Production and Comprehension

Finding the balance between accessibility (with such associated notions as grounding, topicality, and processing ease) and the need to make one’s point quickly in the rapid to and fro of dialogue (with associated notions like focality and attention) has been the subject of increasing interest in more recent psycholinguistic work. Both in theoretical linguistics and in psycholinguistics, however, there has been a tendency to concentrate on the single language user, doubtless because of the long-established emphasis on language competence and the associated individual cognitive abilities – as well as the logistical difficulties tied to experimentation with multiple subjects in conversation. Reference ChomskyChomsky (1965: 3) famously stated that “linguistic theory is concerned with an ideal speaker-listener in a completely homogeneous speech community,” and despite taking issue with this characterization, many functional approaches have adopted a similar orientation to the single language user, as when Reference 158Hengeveld and MackenzieHengeveld and Mackenzie (2008: 1–2) present FDG as “a model of grammar [that] will be more effective the more its organization resembles language processing in the individual” (my emphasis). Reference Pickering and GarrodPickering and Garrod (2021: 11) point out that the very same applies to the cognitive sciences, including psycholinguistics in general, which “have focused on individual rather than joint activities. Specifically, there has been little concern with the mental representations and processes that take place within individuals as they participate in joint activities … This is the case both for joint activity in general and for dialogue in particular.”

Nevertheless, there have been repeated calls for a reorientation to dialogue, and specifically to context-embedded dialogue, in both psycholinguistic and more narrowly linguistic work. This has led to such initiatives in linguistics as Dynamic Syntax (Reference Gregoromichelaki, Kempson, Scott, Clark and CarstonGregoromichelaki and Kempson 2019: 201), which adopts a view of syntactic structure “where utterances are taken as goal-directed physical actions coordinating with equally goal-directed mental actions within and across speakers … such actions are aimed to locally and incrementally alter the affordances of the internal/external context for both one’s self and one’s interlocutors.” The essence of Gregoromichelaki and Kempson’s approach is “a grammar architecture whose core notion is incremental interpretation of word sequences (comprehension) or linearisation of contents (production) relative to context” (p. 192), and their work features such “compound contributions” as (9):

1. (9) A [seeing B emerge from a smoke-filled kitchen]: Did you burn

   B [interrupting]: myself ? No, fortunately not.

where only the dialogic co-construction of the shared (strictly speaking ungrammatical) utterance Did you burn myself? can explain the occurrence of myself rather than yourself (p. 190).

A comparable position is adopted from a more psycholinguistic perspective by Reference Momma and PhillipsMomma and Philips (2018), who argue for a strong measure of similarity between language production and comprehension, specifically with regard to the granularity of incremental processing, that is, the extent to which speakers plan ahead as they utter each unit and the extent to which comprehenders predict upcoming material. The particularities of the language being spoken clearly can interfere with what they call the “left-to-right” planning of lexical items in production (Reference Momma and PhillipsMomma and Philips 2018: 246). In a language with gender like German, for example, it seems likely that the (later) head noun will be planned before any associated definite determiner, the form of which is dependent upon the gender of the noun, whereas this is not a relevant factor in English where the definite determiner is invariable: cf. der Aufsatz (‘article,’ masculine) / das Buch (‘book,’ neuter) vs. the article / the book. Similarly, in comprehension, the gender of the definite article in German narrows the expectations of the hearer and, especially in a context that already limits plausible predictions (say, a discussion of academic publications), contributes to efficient processing. In incrementally decoding the initial words of a question Darf ich deinen … ‘May I [VERB] your …,’ the hearer can discount Buch and entertain Aufsatz as a possible continuation, because of the masculine-accusative marking of deinen, restricting his search space. This greater efficiency is bought, of course, at the expense of remembering the gender of each noun, a long-term memory requirement not imposed on the speaker of English.

Much relevant work in psycholinguistics, understood as the study of how language is produced and comprehended, is oriented to laying bare what is involved in those processes. Accordingly, a focus on the timecourse of verbal interaction entails for many psycholinguists (e.g., Reference Chater, McCauley and ChristiansenChater, McCauley, and Christiansen 2016) that there is no need to posit static and complete knowledge representations (forming a structured and autonomous language competence) that are independent of procedures for language processing, since these are seen as skills that are in large measure comparable with other time-bound skills. The challenge is then to understand the mechanisms of those skills as against the limited capacity of working memory. The solution again lies, in ways that are reminiscent of the division of labor proposed in theoretical linguistics, in striking a balance between sequence and hierarchy, whereby successive units are recoded in comprehension into hierarchically higher units through a process, to use a term familiar from Section 6.3, of “chunking.” The resultant chunks are stored in an inventory that Reference Christiansen and ChaterChristiansen and Chater (2016: 115) have dubbed a “chunkatory,” using an impromptu term that mimics the informal nature of the chunks that are assembled in this way. These are induced from the processes of comprehension and then immediately become available for production processes, that is, implementable as higher-order units available for incremental sequencing. The resultant sequences themselves form ever larger chunks, ultimately yielding a hierarchy or layered network of chunks of different degrees of inner complexity. There are strong parallels between this so-called Chunk-and-Pass approach to production and comprehension (also often referred to as “cascaded processing”) and the proposals of the usage-based work in Construction Grammar (Reference GoldbergGoldberg 1995), where the structured network representing all of the linguistic units we know – of whatever complexity – is referred to as the “constructicon.”

Whereas the Construction Grammar tradition focuses on the unification of chunks to form ever more complex signs, i.e., pairings of form and meaning, the psycholinguistic approach is less concerned with building up semantic representations of formal units and more with how they are interpreted, i.e., how they impact on the hearer’s general understanding of what is being said. These processes of interpretation are triggered by the words said but are only very partially linguistic and involve multiple pragmatic considerations such as the identification of referents or the differential status of information units as “new” or “given” and crucially also a range of contextual factors, which can include world knowledge of various degrees of generality or specificity, the visual context, the social context, and a wealth of prior experience of language chunks and of how they have been interpreted in the past. In interpreting linguistic input, the hearer is involved in a very rapid process of updating, using all imaginable aspects of context to fit the incoming information into a consistent and relevant understanding of the ongoing discourse.

It is becoming increasingly clear that these interpretation processes have to contend with a massive amount of ambiguity in the linguistic input. Reference Piantadosi, Tily and GibsonPiantadosi et al. (2012) have not only demonstrated the ubiquity of multi-interpretable chunks of language but also argued convincingly that the lack of bijective relations between form and meaning is actually “a desirable property of communication systems” (p. 281) because it keeps utterances short and simple. The apparent shortcomings of the signal are more than adequately compensated for by the fact that context is so richly informative about meaning. Hearers draw on context to infer the intended meaning: Any extra cost that this may entail (typically in the form of abductive reasoning) is fully offset by the benefits of having an efficient system that makes copious reuse of the same resources. As Reference Piantadosi, Tily and GibsonPiantadosi et al. (2012: 283) remark, “when the individual units of [such] an efficient communication system are viewed out of their typical contexts, they will look ambiguous”; however, it is only in linguists’ abstract examples that the appearance of ambiguity is troubling. One famous case of such ambiguous-when-out-of-context sentences is the kind of “garden-path” sentences created by researchers with the intention of leading experimental subjects down a syntactic garden path (the reference here is to an idiom that means “cause to take the wrong route”). The prediction is that because of the constructed ambiguity the subjects, confronted with such a sentence out of context, will parse it in the most obvious way, then discover that their parse leads to a dead-end, backtrack, and finally parse the sentence successfully at the second attempt. A familiar example is The horse raced past the barn fell, where raced is initially parsed as a finite verb in the past tense only for the reader to encounter another finite verb in the past tense (fell): [[The horse]_NP [raced past the barn]_VP [fell]_VP]. The reader then tries again, with raced now as a past participle, and the result is a successful parse: [[The horse raced past the barn]_NP [fell]_VP]. However, examination of the use of such garden-path sentences in context (Reference FerreiraFerreira 2008) has shown that they rarely cause any difficulties because the reader who has already encountered “the horse” (definite and specific) will then typically also know whether the horse was being ridden, which would favor the second reading, or running free, which would favor the first. Even the notion of the “most obvious” way to parse a sentence makes an implicit reference to imagined contexts: Since general knowledge tells us that horses are swift animals, it is natural to think of one “racing” and to try that interpretation first.

It has been argued that insights into how the pieces of incoming information are integrated into the hearer’s understanding of the ongoing discourse (the Chunk-and-Pass account) are insufficient to account for the astounding rapidity with which speech is processed and with which speakers initiate their turns. Those insights have been supplemented in recent work by a realization that hearers additionally use probabilistic cues to predict future words and chunks, and even how these will be interpreted, massively augmenting the efficiency of language processing. Reference RappeRappe (2019) refers to this perspective as the Predictive Processing framework. It shares the notion that the hearer converts a linear signal through chunking into a hierarchical representation but differs in making crucial reference to context in the sense of prioritizing higher-level representations, i.e., those that contextualize the ongoing discourse. These relatively general representations become the basis for the hearer’s predictions of what is to come; as Reference RappeRappe (2019: 362) observes, “[l]inguistic processing is at least minimally predictive in that linguistic stimuli come with some sort of context, even if what we call context is only the previously processed linguistic information. This information reaches and affects the system before the input becomes available and facilitates further processing.” If the context is already activated before any linguistic input arrives, it will only be further enriched once the speech begins to flow. One of the options considered by Rappe is that since the aim of understanding is to achieve a satisfactory interpretation of the discourse as a whole, lower-level representations may not be fully constructed in practice (however much linguists may squabble about the details). This is the notion of “good-enough processing” as first presented by Reference Ferreira, Bailey and FerraroFerreira et al. (2002), who found that experimental subjects, confronted with out-of-context sentences, regularly come up with interpretations that are at odds with the complete semantic representation of those sentences. Rather than analyzing the meaning in the manner of a semanticist, the language user employs expectations about the discourse context to predict what a sentence will mean and may then also persist with that interpretation despite linguistic evidence to the contrary. In other words, hearers are found to sidestep strict one-by-one incremental build-up of the meaning of sentences and to commit early to higher-level representations.

The human brain as a predictive organ is central to Reference ClarkClark’s (2015) proposals for the processing of perceptual information. Rather than a one-way flow of sensory information from the external world to the brain, Clark posits sensory information as a potential corrective to top-down predictions: Incoming stimuli then either confirm the predictions or adjust them for future occasions. The top-down predictions can be so strong as to cause the brain to ignore or “explain away” contradictory information. Reference RappeRappe (2019: 374) received a personal communication from Clark in 2018, speculating that “language has evolved specifically to ‘fit’ our multi-level predictive brains.”

6.5 Dialogue

The assumption behind much of the work summarized in the preceding sections is that the human being is inherently social and interactive and that the primary arena for all communication, whether by speech or not, is dialogic interaction in context (Reference Weigand and WeigandWeigand 2017). The grammatical framework to which most attention has been devoted in this chapter, Functional Discourse Grammar, admittedly espouses an architecture that, in line with much grammatical tradition, portrays the linguistic activity of the individual speaker, but that grammar is only part of an overall model of verbal interaction, which assumes the co-presence of an addressee and a “contextual component” to which both dialogists have access and which represents the common ground that is the basis for dialogue. Certain aspects of this context are relatively time-stable (e.g., institutional status and roles, or identities in terms of seniority, gender, sexual orientation, ethnicity, physical ability), while others are in continual flux, reacting to the vicissitudes of the moment, to the continual updates arising from the ongoing interchange (Reference Hengeveld and MackenzieHengeveld and Mackenzie 2014; Reference MackenzieMackenzie 2014b). For a thorough-going consideration of how common ground is constructed in the process of interaction as a “shared workspace,” see Reference Pickering and GarrodPickering and Garrod (2021).

Where the context is relatively stable and mutually understood, communication can in principle take place without any use of language, even between individuals capable of language use: Pointing, other gestures, eye movements, etc. may function as discourse acts and be enough to guarantee successful interaction. There are also specific situations in which silent communication is indeed favored, for example, in hunting to avoid being heard by the prey or in secret gestural behind-the-back communication between doubles partners in a tennis competition, or where the background noise is so intense that gestures are the only way to communicate information. The amount of information that can be imparted in this manner is limited, of course, in the first instance to indexical identification of some aspect of the physical context and human movement within that context. But the dependence upon context is extreme in such communication and it is prone to misunderstandings. The challenge may lie in distinguishing between gestures that arise from a communicative intention and other chance bodily movements that are not designed to convey information.

However, there are many situations in which it is perfectly possible for adults who are capable of speech to cooperate without talking. As the Viennese linguist Reference BühlerBühler (1990: 47) pointed out, “Productive and creative man, working in cooperation with his fellows, often remains silent as long as each fully understands what the other is doing and acts appropriately.” Such situations can be said to exploit the context of communication to the maximum: Where there is a high measure of common ground, speech is redundant and indeed superfluous. Reference BühlerBühler (1990: 176) asks, “To what end should one speak if one gets on just as well or better in practical life without speaking?” He answers by invoking “islands of language” emerging “from within the sea of silent but unequivocal communication at the places where a differentiation, a diacrisis, a decision between several possibilities has to be made, and easily can be made by interspersing a word,” exemplifying this statement with his memories of earlier times in Vienna when there was a single kind of tram ticket and silent interaction with the conductor sufficed; only with the introduction of simple and transfer tickets did the need for the “empractical” (i.e., stand-alone) utterances Geradeaus “straight through” and Umsteigen “change” arise. The holophrastic utterances in question are, it should be noted, no more complex than is required in the highly restricted framework of two public transport options. One is an adverb, and the other is a verb in the infinitive form, but each is complete in its context; no misunderstanding is possible.

Of course, the major emphasis in studies of dialogic interaction has been on less constrained situations, where the syntax is more sophisticated and the full multimodal mix is available (notably prosody and gesture). The dialogists have a sufficiently shared awareness of the context and sufficient experience of how it interacts with the linguistic cues on offer to make communication in most situations satisfactory for both participants. As has been mentioned in previous sections, language is both an imperfect tool, in that the relation between communicative intentions and linguistic forms is complex and insecure, and also a close-to-perfect tool in making creative use of a restricted set of syntactic and lexical instruments. The result is dialogue as a form of tacit cooperation (even in the case of a furious row between the interlocutors), in which throughout each utterance the operations of chunk-and-pass and of predictive processing actively contribute to the inferences that each is drawing about the other’s communicative intentions. Understanding is never guaranteed, however, and conversations can jump to the meta-level, with requests being made and granted for reformulations, before the conversation returns to the question under discussion.

The dialogical perspective upon language use draws attention away from the criterion of grammaticality toward that of situation-appropriateness (Reference Anward and LinellAnward and Linell 2016). From the viewpoint of the participants in the dialogue, the ultimate questions are: How much does the utterance contribute to the achievement of our goals as interactants? And how well does it fit into the context in which we find ourselves and which we are co-creating? Prioritizing answers to these questions leads Anward and Linell, in their study of dialogues in Swedish, to make observations that are unattainable with a methodology that considers only sentences in isolation.

One is the ubiquity of “responsiveness,” i.e., structures designed to occur in reactive moves, as in (10):

1. (10)

   a.	å	vilken	hjälp	har	du	fått?
   	and	what	help	have	you	received
   	‘And what help did you get?’
   b.	inte	jättemycke		hjälp	har	ja	inte	fått.
   	neg	awfully.much		help	have	I	neg	received
   	‘Not awfully much help did I get.’

The structure with the double negative shown in (10b) only occurs in responses: The initial focus “not awfully much” morphs into a topic “awfully much” with its comment “I did not receive.” The initially placed focus serves the dialogic purpose of answering the question promptly (see the discussion of Hannay’s “reaction mode” in Section 6.3 above), while the topic structure with its reuse of hjälp and fått contributes to the alignment of question and answer. Another property found in the Swedish dialogues is “incrementality,” understood as we have seen as the stepwise build-up of an utterance, with the speaker not knowing exactly how it is going to end; it may even “end up in an incoherent syntactic structure, as judged from conventional sentence grammar” (Reference Anward and LinellAnward and Linell 2016: 17; emphasis in original), but without triggering any reformulations or requests for repair. A final property is “syntactic ambiguity,” understood here as referring to multiple internal relations that arise from changes in pragmatic properties in the course of the utterance, as in (11), which is not “grammatical” as it stands but shows evidence of its piecemeal construction in real time:

1. (11)

   de	va	bra	(.)	att	han	gjorde.
   it	was	good	pause	that	he	did
   ‘It was good that he did it.’

In this case, the pronoun de is initially understood as anaphoric; then the delayed incrementation of the “extraposed” subordinate clause renders it cataphoric; and finally, it comes to be interpreted as the object of gjorde. This example shows the dynamic interaction between linguistic form and context: The speaker initially assumes contextual clarification of the implicit reference of de; then after a pause makes this explicit but in the meantime has reinterpreted the de as the topic of the utterance and yet semantically integral to the subordinate clause, which consequently lacks its full complement of arguments.

These examples manifest a usage-based view of linguistic data, presented in the form of transcriptions of speech, as protocols of joint action that took place between certain individuals at a certain time in the past. The analyst can discern, in the left-to-right sequencing of data, the effects of those individuals’ incremental co-creation of meaning. However suggestive such linguistic data may be, the need for objective criteria has encouraged the use of various new technologies designed to detect the cognitive processing underlying dialogical speech. It has long been understood, relatively informally, that such matters as speech-accompanying gestures, changes in interactants’ direction of gaze, interruptions, and requests for speech repairs can all be treated as clues to what is going on, but it is only fairly recently that equipment for tracking saccades (rapid movements of the eye between fixation points) has become available, allowing precise conclusions to be drawn about how dialogists interpret each other’s disfluencies and prosody. In particular, use is made in laboratory settings of “confederates,” that is, members of the team who enter into dialogue with experimental subjects who are fitted up with the gaze-tracking equipment. The confederates work with scripts that maneuver the naïve subjects toward certain ambiguities, allowing the researchers to infer relevant disambiguating cognitive processes from the subjects’ eye movements (see Reference Brennan, Hanna and WeigandBrennan and Hanna 2017 for details). It is expected that future developments will reinforce our understanding of the sequences of cognitive processes that underlie the incremental build-up of discourse acts and longer stretches of discourse.

6.6 Conclusion

This chapter has offered an overview of developments that reflect a convergence of work in linguistics and psycholinguistics around the implications of the piecemeal composition of units of speech for an understanding of grammar and the cognitive processing that underlies the production, comprehension, and interpretation of utterances. Notions from Functional Discourse Grammar have been employed to present a view of syntactic structure as arising from the incremental expansion of the minimal utterance, the holophrase. By prioritizing the timecourse of language processing, we have interpreted syntactic hierarchy as arising from chunk-and-pass operations, arguably supported by predictive processing. Finally, our attention turned briefly to dialogue as a general framework for the understanding of interaction, communication, and language. All of these notions make essential reference to context as a necessary component of the operations and processes that we have reviewed.

7.1 Introduction

Cognitive Linguistics, as meant in this chapter, is the linguistic framework represented by handbooks like Reference Evans and GreenEvans and Green (2006), Reference Geeraerts and CuyckensGeeraerts and Cuyckens (2007), Reference Littlemore and TaylorLittlemore and Taylor (2014), Reference Dąbrowska and DivjakDąbrowska and Divjak (2015), Reference DancygierDancygier (2017), or Reference Wen and TaylorWen and Taylor (2021), readers like Reference Evans, Bergen and ZinkenEvans et al. (2007) or Reference GeeraertsGeeraerts (2006), and shorter introductory texts like Reference Winters and NathanWinters and Nathan (2020). The very fact that I am identifying the approach by means of encompassing reference works and not by referring to a single dominant name is significant. Generative grammar is predominantly and perhaps overwhelmingly Chomskyan grammar, just like Systemic Functional Grammar is Hallidayan linguistics, but Cognitive Linguistics is not the linguistic approach developed by a single Big Name. Rather, like perhaps American structuralism or the Prague School, it is best characterized as a movement fed by the ideas of a number of inspirational figures. In the first generation of cognitive linguists, these would include people like Ronald Langacker, George Lakoff, Leonard Talmy, Gilles Fauconnier, with other key figures emerging in later generations. The recognition that Cognitive Linguistics is a collective enterprise (and perhaps a rather loose one at that) raises the question of what holds the framework together: There will surely be an affinity between the various ideas introduced by the central members, but how can that affinity be best described? A network of thinkers produces a network of analytical concepts and descriptive practices, but what are the cohesive forces within that network? In this chapter, I will develop the idea that context is a crucial factor not only for the internal cohesion of Cognitive Linguistics but also for the position of Cognitive Linguistics in the evolution of linguistics in the twentieth and twenty-first centuries. In particular, I will argue that Cognitive Linguistics is a recontextualizing framework par excellence, i.e., one that reintroduces the contextual embeddedness of language in linguistic theory formation and that in doing so it counterbalances the decontextualizing tendencies that are at work in formal grammar and formal semantics.

Three caveats will be useful before we proceed. First, emphasizing the importance of context for Cognitive Linguistics does not imply that it is the only contemporary or recent framework in linguistics in which context plays a central role. In particular, much of functional linguistics is about contextual factors and language as a social semiotic with a communicative function. However, a systematic comparison between functional and Cognitive Linguistics (as in Reference Nuyts, Geeraerts and CuyckensNuyts 2007) is beyond the scope of the present chapter. Second, as implied by the previous paragraph, I will try to do justice to Cognitive Linguistics as a diverse framework that brings together different, largely complementary approaches to language, without privileging any one in particular. Specifically, although some readers may tend to associate Cognitive Linguistics predominantly with the Lakoffian tradition of research into conceptual metaphors and embodied cognition, they should be aware that this line of investigation, however impressive, is but one strand within the larger texture of Cognitive Linguistics. Third, I have made the point about the recontextualizing nature of Cognitive Linguistics before (Reference Geeraerts, Tabakowska, Choinski and WiraszkaGeeraerts 2010, Reference Geeraerts2015), and some readers might remark that the following pages are basically a recombination of ideas that I expressed earlier. They would not be wrong, but even if the present text will repeat, to some extent literally, my earlier treatments of the issue, it will also offer a more analytical and systematic view of the recontextualizing core of Cognitive Linguistics.

7.2 Cognitive Linguistics in the History of Modern Linguistics

To understand the specific position of Cognitive Linguistics in the history of modern linguistics, we may focus on two aspects of generative grammar: its changing view on the role of semantics, and its overall decontextualization of grammar as autonomous syntax.

7.3 Recontextualizing Meaning

To appreciate how Cognitive Linguistics recontextualizes meaning, we will first go over the most prominent models for meaning description suggested by Cognitive Linguistics, and then identify the two major ways in which these approaches embody a contextualized perspective on meaning.

7.4 Recontextualizing Grammar

In the previous section, we saw not only how Cognitive Linguistics places semantics at the heart of linguistic description, but how it specifically develops a contextualized conception of meaning. The basic impulse of Cognitive Linguistics is to put the description of meaning in a central position in linguistics, but just as importantly, that meaning is not studied as something isolated but rather as something that is embedded in a wider context – and that context takes different forms: the body, society and culture, and actual language use. In this section, we will see how that recontextualizing tendency also applies to the way in which Cognitive Linguistics looks at linguistic systems at large. As in the previous section, there are two relevant features to pay attention to, and they are in fact related to the features discussed there. First, similar to the interactive relationship between concepts and their contexts of usage, we will describe how usage-based linguistics generalizes that idea to language at large. Second, similar to the experiential grounding of meaning, we will introduce the intersubjective grounding of linguistic systems as the basis for a “social turn” in Cognitive Linguistics.

7.5 Conclusion

Let’s wrap up. We have positioned Cognitive Linguistics against the background of twentieth-century developments in linguistics and described how Cognitive Linguistics constitutes a recontextualization of linguistics in contrast with formal semantics and formal grammar. In contrast with the former, Cognitive Linguistics works with a model of meaning that emphasizes the experiential grounding of meaning and the dialectic interaction of senses and context of use. In contrast with the latter, Cognitive Linguistics develops a usage-based model of linguistics and reintroduces the social dimension as an inherent dimension of language, and social variation as an inherent part of grammar. Both pairs of recontextualizing trends are related, as can be readily appreciated if we interpret them against the background of the semiotic triangle as classically described by Reference Ogden and RichardsOgden and Richards (1923). The semiotic triangle relates the linguistic sign, on the one hand, to its reference, and on the other, to the language user. These two poles may precisely provide a link between the two pairs of recontextualizations, if we double the triangle by not only applying it in its original form to individual signs, but by also applying it to language at large. In the latter case, we consider how the linguistic system (the large-scale equivalent of the individual sign in the basic Ogden and Richards conception) relates to users and groups of users, and then to its instantiation as usage events. “Instantiation” is the crucial notion here: Just like the referents or exemplars of a single sign instantiate the associated category, a linguistic system is extensionally represented by all the occasions on which, and all the specific forms in which, it is put into practice.

If we accept this doubling of the semiotic triangle (corresponding, needless to say, with the distinction between Sections 7.3 and 7.4 in the pages above), we can define the essential similarity between the two forms of contextualization that we identified within each of the two perspectives. First, both the experiential grounding of meaning and the intrinsically social, intersubjective nature of language involves the user: users sharing common physiological and anatomical embodiment; users with their individual experiences and background; users as part of intersubjective and communicative communities. Second, a dialectic relation between language and its instantiation in a specific situational context exists both for individual linguistic categories and for language as a whole. The resulting picture of the fundamental recontextualizing characteristics of Cognitive Linguistics in their mutual relations can be summarized in Table 7.1.

It appears that (re)contextualization indeed provides a unifying perspective on the contribution of Cognitive Linguistics to the development of linguistic theory in the late twentieth and the early twenty-first century. But to round off, it should be remarked that this unifying perspective should not be taken to imply a complete homogeneity of Cognitive Linguistics. The systematic and pervasive relevance of recontextualization within Cognitive Linguistics should not obscure the fact that tensions may exist within the framework: Even if recontextualization is a defining overarching characteristic of Cognitive Linguistics, various forms of recontextualization do not necessarily point in the same direction. Specifically, an experiential grounding of meaning as embodiment has a universalist tendency (assuming that the human body is the same for everybody), while a sociocultural conception of language will tend to emphasize the diversity and historicity of the human experience. If it is correct to characterize Cognitive Linguistics in terms of recontextualization, determining the proper relationship between what is universal and what is historical and specific may well be a major focus for the further evolution of the framework.