What is a Context?

Abstract

In this chapter, I examine a variety of criteria for context individuation, with an eye towards identifying and distinguishing various theoretical projects in semantics and pragmatics for which they would be appropriate. The discussion is organized around six possible desiderata for contexts: (1) that they metaphysically determine proposition expressed, (2) that they be non-trivial (in a sense explained in the text), (3) that they provide epistemic illumination of how hearers understand what proposition was expressed, (4) that they be (in various senses) finite, (5) that they be non-intentionally characterized, and (6) that they be context-insensitively characterized. Particular attention is directed to the conceptions of context developed by Lewis, Stalnaker, and Davidson(ians).

This paper originally appeared as Chapter II of my 1998 Harvard Ph.D. Dissertation, which was subsequently published as Gross (1998/2001). Many thanks to Alessandro Capone for the invitation to reprint it here.

1 Introduction

Language users occasionally utter sentences. In attempting to explain this behavior, theorists about language often advert to various features of these occasions under the rubric: the context of utterance. What is a context?

Let’s narrow our scope somewhat. Sentences are uttered as parts of various speech-acts, and features of the occasion may be invoked to explain what determines which speech-act is performed. Suppose, though, we just assume it given that the speech-act in question is assertion. Now, a typical instance of assertion involves (1) an utterance of a sentence with a certain standing meaning in a language, in order thereby (2) to express a proposition, and perhaps in addition (3) to communicate pragmatically some further proposition(s). Features of the occasion of utterance may be invoked in explaining facts at any of these three levels—the pre-semantic, semantic, and post-semantic, as they might be called.¹ But suppose we focus just on level (2): what determines what proposition is expressed, given the meaning of the sentence uttered. (It’s clear that natural language sentences are pervasively context-sensitive in this sense: what propositions speakers express in uttering them is not determined by their meaning alone. ‘The kettle is black’ can be used to express a variety of propositions, depending for instance on the kettle in question, the time of utterance, whether the layer of soot matters (the speakers might be photographers composing a black-and-white still-life, or they might be kettle collectors wondering whether, beneath its sooty exterior, this is a rare white 1924 model XYZ), and more.) Our focus thus narrowed, how ought we to characterize and individuate contexts?

The primary point of this chapter is to argue that this is a bad question—at least until we specify what we want a notion of context for. Contexts are just (representations of) bundles of facts. There’s no sense to inquiring into their proper individuation conditions without an idea of what questions we hope to answer in part through adverting to context. Only given an idea of the task to which they will be put can we debate what information they should encode. We thus can say what a context is only after clarifying our theoretical aims.

Our preliminary narrowing has already illustrated how context may be invoked to help explain different aspects of linguistic behavior. In what follows, we shall see that even after this narrowing there remains room for different notions of context, corresponding to different questions one might want answered. Since the proper characterization of context depends on one’s particular explanatory ends, certain debates about context lose their point: conceptions of context are not in competition, if they are meant to serve different ends. Showing this is fruitful not only because it helps clarify—indeed dissolve—certain disputes, but also because, in forcing a clarification of these varying explanatory ends, it elicits a sharper delineation of the varying projects aimed at those ends.²

I begin by listing six desiderata one might want an account of context to satisfy.³ I then turn to three accounts of context (identified with Lewis, Stalnaker, and Davidson) to see how they fare. (Assessing the theoretical potential of these three conceptions of context in particular is thus a secondary aim of this chapter.) What will be crucial to note is that whether these desiderata ought to be accepted—whether they place legitimate constraints on an account of context—depends on what work one’s notion of context is supposed to do. Without a clear conception of this, one cannot assess whether a failure to satisfy some desideratum should count as an objection. We shall see that the three conceptions of context are addressed to three different theoretical tasks, which require satisfying three different sets of desiderata.

2 Six Desiderata

The first desideratum is the minimal demand that the context, given the meaning of the sentence, determine or fix the proposition expressed. What was said ought to supervene on context plus sentence meaning. If this weren’t the case, then one and the same sentence, with the same meaning, could be uttered in one and the same context, and yet express a different proposition. But then, presumably, there was some further factor that in part determined what proposition was in fact expressed. We should then include this factor—or something that determines it—in the context. That is, we want contexts to be (if you will) complete with respect to the task of determining the proposition expressed, given the meaning of the sentence uttered. I say that this is the minimal demand on contexts, because it simply amounts to a restatement of the task on which I suggested above we focus our sights.

Second, we might require that contexts not be trivial. Contexts are trivial if they are characterized in a way that begs the questions we want to answer. For example, a trivial way of satisfying the demand for determination is to characterize contexts in terms of what proposition is expressed. What proposition is expressed will of course supervene on itself, but what is wanted presumably is some characterization of propositional determination in independent terms. Perhaps it could be argued that no such account is forthcoming, at least none meeting other suitable constraints. But until such an argument is compellingly made, the explanatory aspiration embodied in the first desideratum will seem reasonable and worthy of pursuit. Similarly for whatever other explanatory aspirations we might have: if there is some question we want answered, we will want it answered, if possible, in a non-trivial manner.

Third, we might require that contexts be characterized in a manner that is epistemically illuminating. For example, we might require that the features constituting context be among those language users actually track in understanding one another—or, at least, that they be such that they could in principle enable such understanding, given our cognitive capacities. This would require both that the features be epistemically accessible and that their bearing on propositional content, given sentence-meaning, be apparent (at least upon reflection) to normal speakers of the language. Suppose what proposition is expressed on any given occasion happens to supervene on some complex microphysiological state of the speaker. Contexts comprising just this state would satisfy the determination requirement in an epistemically unilluminating way, since this state would neither be (in general) epistemically accessible to hearers capable of understanding what is said nor such that those hearers would or could (in general) grasp its bearing on content (even if the state were epistemically accessible to them).

Fourth, we might require that contexts be finite, that they comprise a finite number of features. (If contexts are represented as n-tuples, the requirement would thus be that n be finite.) This requirement could stem from various sources. For instance, it might arise as a constraint imposed by the form semantic theorizing might take. If our task is to articulate all the ways contextual features may affect propositional content, its completion might require finiteness.⁴ Or a finiteness constraint might be motivated by learnability considerations.

Now, it may be noted that finiteness seems all too easy to satisfy: one can always replace an infinite sequence with a finite one that contains one object—viz., the infinite sequence itself. So, we might want rather to require that contexts be finitely grounded, that they comprise a finite number of features that themselves comprise a finite number of features, and so on for a finite number of levels. There are two other relevant refinements perhaps worth mentioning here. One might want to require that the number of features be not only finite, but also fixed, so that no change in the set of relevant utterances (e.g., in anticipation of possible changes in the language) would require the addition of contextual features. Further, one might require not only that contexts be finite, or finitely grounded, but in addition that they be manageable—that is, that the list of features be reasonably short. A finite, but utterly unwieldy list might be practically useless, at least for some purposes; and its lack of perspicuousness might render the account of which it is a part unilluminating.⁵

Fifth, we might require that contexts be characterized in non-intentional terms. Like the finiteness requirement, this constraint might have a variety of motivations. For instance, one’s semantic theorizing might be intended to subserve a broader reductionist project. Intentional characterizations will be precluded if one’s naturalizing strategy, for example, is to reduce the mental to the linguistic and then the linguistic to the non-intentional.⁶ Or the requirement might stem from a desire to advert only to relatively epistemically uncomplicated contextual features—in particular, features not subject to the epistemological holism of the mental. Finally, at least in the case of certain intentional terms, the constraint may simply follow from the previously mentioned fear of triviality, as we shall see below.

The last possible desideratum I will mention is that the characterization of context be itself context-insensitive. The worry behind this constraint is that a theory that adverts to context-sensitively characterized contexts will itself be context-sensitive, so that what the theory itself says will vary across contexts—or, rather, there won’t be any one theory put forward in these terms. It doesn’t follow just from this that a context-sensitive characterization of context precludes a general theory, one that is about utterances and contexts generally. For perhaps from a distinguished context (the contribution to content of context-sensitive terms fixed by that context), a theory can be put forth that achieves this generality. But if so, then it is straightforward to supply context-insensitive replacements—e.g., by stipulating, as a new technical sense for the terms, that they context-insensitively contribute to content what the terms in their non-technical sense context-sensitively contribute in the distinguished context. This desideratum, then, requires that this—or something equivalent—can be done.⁷

3 Three Conceptions of Context

Let me now introduce three conceptions of context found in the literature.⁸ The first is due to David Lewis, the second to Robert Stalnaker. The third—perhaps not due to any one person in particular—is the notion of context required by what I consider the best version of (neo-)Davidsonian semantics. The Lewis and Stalnaker conceptions of context are easily introduced. Getting the Davidsonian conception on the table will require a few words about his semantic program.

A context, for Lewis, is an ordered triple: the “location—time, place, possible world—where a sentence is said.”⁹ According to Stalnaker, a context is a speaker’s presuppositions, “the propositions whose truth he takes for granted as part of the background of the conversation.”¹⁰ Stalnaker favors representing such a context as the set of possible worlds compatible with what is presupposed; but this is a further step, independent of the basic conception of context.

On the Davidsonian conception, a context comprises the features adverted to by a truth-theory that assigns truth-conditions for the sentences of a given language.¹¹ These truth-conditions are captured by “T-sentences” derivable within the theory. For context-sensitive sentences, which possess variable truth-conditions, such T-sentences take the following “normal form”¹²:

(u)(x)(y)… (z)[If uR(S, x, y,…, z), then {u is true iff F(x, y,…, z)}].

(In something closer to English: For any utterance of a sentence S and any contextual features x, y,…, and z, if those features obtain on the occasion of utterance (and meet whatever other relevant constraints there might be), the utterance is true if and only if x, y,…, and z satisfy F.) The features constituting a context would be the values of variables x, y,…, and z. For example, consider the context-sensitivity arising from the variable reference of ‘it’ in ‘It’s green’ (ignore other sources of context-sensitivity). Its T-sentence would be:

(u)(x)If u is an utterance of ‘It’s green’ in which ‘it’ refers to x, then (u is true iff x is green).

The T-sentence requires that contexts be at least individuated according to what object is referred to by ‘it’ on that occasion. The T-theory as a whole commits one to individuating contexts according to whatever features are thus adverted to within it.¹³

How well do these three conceptions of context fare in satisfying the various desiderata listed above?¹⁴

• (a) Lewis

Lewis’ conception of context as location clearly satisfies desiderata 1, 2, 4, and 5. What proposition is expressed certainly supervenes on the utterance’s having occurred at that place, time, and world¹⁵; these three features are clearly non-trivial—at least so far as content-determination goes—in that they don’t invoke what requires explanation; they are finite in number, finitely grounded, presumably fixed, and, being only three, manageable; and they are obviously non-intentional.

Whether desideratum 6 is satisfied—whether such contexts can be context-insensitively characterized—is, however, not so obvious. There are two issues here. First, specifications of place and time are context-sensitive as to their relevant extent: how extensive are the place and time that in part constitute the speaker’s location? I won’t dwell on this, however; perhaps a reasonable stipulation will settle the matter. Second, it is unclear whether we can context-insensitively refer to an arbitrary world. Location in space–time does not pose this problem, since we may avail ourselves of a coordinate system (dubbed with a name—once an origin, orientation, and unit magnitude are stipulated¹⁶). But context-insensitively picking out a location in the space of possible worlds is not so easy. We may provide our world the context-insensitive name ‘@,’ and likewise take care of some others we have first context-sensitively picked out. But we seem to lack the means to refer context-insensitively to an arbitrary world.¹⁷

But is this a problem? Lewis has, after all, context-insensitively characterized what contexts are (supposing the first concern addressed): they are places, times, and worlds.¹⁸ This suffices to individuate contexts, even if we lack the means context-insensitively to refer to arbitrary contexts. This second problem, then, is really only a problem if there is some reason Lewis need be committed to our having such an ability. We thus see that there is a distinction to be drawn between context-insensitively characterizing what contexts are and context-insensitively characterizing (arbitrary) particular contexts in a manner sufficient to single them out. Identifying what suffices to determine propositional content (given sentence-meaning) requires only the first. Even if your aspirations include articulating a theory that enables the context-insensitive specification, for an arbitrary utterance in a context, of the proposition it would express, this in fact would suffice. Articulating the theory doesn’t require the ability to refer context-insensitively to arbitrary contexts: it doesn’t even require the ability to refer to arbitrary contexts simpliciter (whether context-sensitively, or context-insensitively). However, if you lack this ability, you’ll only be in a position to apply the theory to utterances entered in contexts to which you could refer.

The failure of Lewis’ conception of context to satisfy desideratum 3—the requirement that contexts be epistemically illuminating—is more obvious. Lewisian location is certainly knowable (minimally, we can demonstratively identify our current spatial, temporal, and modal whereabouts). But from the fact that P supervenes on Q, it doesn’t follow that finite agents, who as it happens can recognize that Q, can thereby learn that P. And indeed knowledge of these features, combined with knowledge of what sentence was uttered with what meaning, does not in general enable normal competent speakers to recover propositional content. Knowing that ‘I love you,’ for instance, was uttered in our world at precisely noon on January 12, 1879 in room 17 of the Dakota off Central Park obviously does not enable me to know what proposition was expressed.¹⁹ Lewis’ conception of context, though it may help answer questions requiring only the determination of propositional content, seems ill-suited to addressing how speakers succeed in understanding what is said.²⁰

But is this a criticism of Lewis? Not if addressing how speakers succeed in understanding what is said just isn’t his aim—or at least not his aim in advancing this conception of context. So what is his aim? Lewis tells us that his advancing this conception of context subserves his attempt to provide “part of a systematic restatement of our common knowledge about our practices of linguistic communication [… that] assign[s] semantic values that determine which sentences are true in which contexts.”²¹ Providing that part that covers our understanding of what others say is simply not the task he sets himself here. Indeed, in other work, he seems to hold that at least part of this task requires something like Stalnaker’s conception of context.²²

• (b) Stalnaker

Let’s turn then to Stalnaker’s conception of context as the set of propositions the speaker presupposes to be mutual among participants in the conversation. The first point to note is that for Stalnakerian contexts generally to satisfy desideratum 1—for them to determine propositional content, given sentence-meaning—it is arguable that an assumption of normality must be made. A context is normal—or, as Stalnaker puts it, “non-defective”—when the speaker’s presuppositions are in fact the same as her audience’s.²³ When the context is defective, it can be that it is rather the hearer’s presupposition set that determines the proposition expressed—at least this is a defensible position in some cases.²⁴ The most likely such cases are ones in which the hearer’s presuppositions are very reasonable, but some of the speaker’s are not. For instance, if Tommy and Suzy are playing with Suzy’s marbles, and Tommy declares “All the red ones are mine,” it is unclear that the fact that he has in mind rather the marbles he left at home should suffice to render them the domain of discourse relevant to the assessment of the proposition expressed (perhaps as opposed to the proposition he meant).²⁵ It is unreasonable of Tommy to expect Suzy to share the relevant background beliefs in this case; Suzy’s most likely assumption—that he is falsely claiming ownership of the perceptually salient red marbles, the ones before them that they’ve been playing with for the past hour—is much more reasonable. Perhaps one should conclude in such cases rather that it is in some sense indeterminate what proposition was expressed.²⁶ Then there would be no objection to maintaining that Stalnakerian context determines propositional content (given sentence-meaning), since cases in which there’s no (determinate) proposition expressed at all would just be irrelevant to the claim. But if one is willing to allow that propositional content can in some cases be determined by what the hearer presupposes, then the claim holds only if normality is assumed.²⁷

To satisfy the second constraint—that the context be non-trivial—the speaker’s presuppositions must not include the proposition that she is asserting that P, since among what wants explaining is that (the hearer can grasp that) this is the case. Of course, the speaker believes that this is what she is doing, but normally she can’t expect the hearer to believe this prior to his grasping what the speaker says.²⁸ Once the assertion is made, it can indeed be reasonably expected that the speaker’s having asserted that P will be mutually presupposed. This will typically be among the effects on context the speaker intends her speech act to have. But this just means that the speaker’s having asserted that P will be conversationally available as background for further assertions. There seems no reason, then, to doubt the non-triviality of Stalnakerian contexts.

Now, in adverting to a speaker’s presuppositions, one obviously violates the non-intentionality constraint—or, rather, what follows is that this conception of context is not suited for projects that motivate this desideratum. This is not to say that Stalnakerian contexts can’t be deployed by someone with naturalistic ambitions, but only that one’s strategy must be of the appropriate sort: Stalnaker himself advocates an approach that would explain linguistic intentionality in terms of mental intentionality, and then the latter in terms of a broadly speaking causal story.²⁹ This involvement with the mental, however, may well limit the epistemic illumination forthcoming from an approach that utilizes such contexts. Before examining this, it will be useful to discuss desideratum 4—the finiteness constraint.

Stalnakerian contexts are trivially finite; for they comprise a single feature—viz., what the speaker presupposes. Here is a case where it’s more interesting to ask instead whether the context is finitely grounded. The proper answer to this question is clouded by questions concerning the proper analyses of tacit belief and common knowledge.³⁰ I certainly act as if I believe that I am under 10 ft tall. Do I tacitly believe it? What about my being under 11, 12 ft tall, etc.? If I do tacitly believe all these things, don’t I also tacitly believe that you believe them, and that you believe that I believe them? So, aren’t they among my presuppositions? If so, then presupposition sets need not be finitely grounded. And even if I don’t have all those tacit beliefs (nor all those presuppositions) about my height, consider some arbitrary P that I do presuppose on some occasion of utterance—for instance, that I am speaking. Well, if it’s a presupposition, then you presuppose it also. And that we both presuppose it will also be included among what we take for granted. So, we’ll each presuppose that we each presuppose it, etc. The mutuality built into presupposition can seem to ensure an indefinite iteration of presuppositions. So, if this proves correct, we must again conclude that presupposition sets are not finitely grounded.

These brief remarks are intended merely to mark the issue, not to convince.³¹ At the very least, however, it would seem that the number of propositions in a presupposition set could be enormously large.³² Does this threaten unmanageability? That depends on what work the theory employing this notion of context is supposed to do. Let’s return to the question of epistemic illumination to see one way this worry could be pressed.

Stalnakerian contexts certainly satisfy the conditions necessary for being epistemically illuminating. It is obvious that the attitudes adverted to are epistemically accessible, and that their bearing on propositional content is apparent to actual speakers. If explaining our ability to understand one another is among our goals, it is clear that presupposition sets to this extent fit the bill in a way that Lewisian locations can’t. But why might this conception of context limit the epistemic illumination approaches employing it may provide? Well, suppose we wanted some sense of how hearers, on the basis of their knowledge of sentence-meaning and context, (could reasonably) form a belief as to the proposition expressed. A standard thought is to try to articulate rules that could take one from sentence-meaning and context to propositions.³³ But now the worry is that too much has been included in our contexts to make this feasible. Sure, not all the speaker’s attitudes are there—but still the presupposition set can seem to comprise such a large, unwieldy, holistically intertwined body of information as to render the task intractable. “He’s crazy,” my conversational partner says. Among the presuppositions of the context are that we have been talking about her husband, that her child is visually salient, that her husband is visually salient, that she is pointing at her child, that a dog just barked, that she is gesturing vaguely with her head over her shoulder, etc. What rule will take me from this mess of facts (and many more) to the conclusion—obvious to anyone minimally competent and present—that, in uttering ‘He’s crazy,’ she referred to her dog (who, by barking, had interrupted her story about her equally crazy husband) while she simultaneously indicated to her husband, by pointing to her child, that he should keep an eye on the baby? Now, merely pointing out the complexity of our cognitive abilities in this fashion of course does not constitute an argument that such rules could not be articulated. But it at least serves to indicate that constructing such a theory would require something approaching a complete theory of rational psychology. We needn’t place bets here on that project. Suffice it to say that if attempts to codify “general intelligence” are subject to in principle limits, then it seems that semantic projects that utilize Stalnakerian contexts could be as well. Semantics pursued in this fashion would then be as hard as pragmatics.

Of course, the proper response might be that this constitutes no objection to Stalnakerian contexts. It merely provides a reason not to employ such contexts in that explanatory project.³⁴—Assuming, that is, that such a project can be carried out at all: there’s even less reason to object if it can’t. If it can, however, or if we at least possess no grounds at present for thinking it can’t, then what conception of context ought one employ? A possible moral to draw from our examination of Lewisian contexts is that there’s no avoiding intentionally characterized contextual features if one aims at the sort of epistemic illumination described above. But now we see that a profligate adversion to mental states threatens intractability. The Davidsonian conception can be viewed as an attempt to find a middle ground suitable to the project of articulating semantic rules. It adverts to intentionally characterized contextual features, but attempts to let in only those directly relevant to the calculation of propositional content. What features of context must be tracked in order to calculate the semantically relevant contextual features themselves is a question, according to this view, that semanticists needn’t address.³⁵ The larger questions of rational psychology can thus be left to one side, if this approach is correct. Before turning to this conception of context, however, it remains briefly to remark upon the context-sensitivity of Stalnakerian contexts.

Again, we must distinguish the question of context-insensitively characterizing what contexts are from that of context-insensitively characterizing particular contexts. Regarding the former, there is, first, the vagueness, and consequent context-sensitivity, of ‘presupposes;’ and, second, the arguably context-sensitive individuation conditions for propositions, the objects of the speakers’ presuppositions—i.e., the context-sensitivity of ‘same proposition as.’ Regarding the latter, there is further the question of whether the contents of others’ attitudes can even in principle be context-insensitively characterized. Many accounts of attitude ascription posit context-sensitivity of one sort or another that could not be easily removed.³⁶ In addition, it is arguable that there are some essentially indexical contents expressible only using context-sensitive means.³⁷ In any case, as we have seen, the violation of finite groundability (or at least manageability) would seem to leave us unable in at least some cases to characterize exhaustively a Stalnakerian context, whether context-insensitively or not.

• (c) Davidson(ians)

The Davidsonian conception of context, recall, is that required by a truth-theory for a language: it comprises the parameters to which the theory must advert in specifying the variable truth-conditions of context-sensitive sentences. The determination and epistemic illumination constraints—desiderata 1 and 3, respectively—will clearly be satisfied if such a theory can be constructed. First, given such a theory, we may take the relevant T-sentence to provide—or at least to be determined by—the uttered sentence’s meaning. Then, if one plugs in the relevant contextual features, the biconditional’s right-hand side will yield what was said.³⁸ Second, a successful theory will advert to epistemically accessible contextual features whose bearing on propositional content is explicitly exhibited in the T-sentence, which would thus seem better to fit the above ideal of a semantical rule. Whether such a theory can indeed be constructed for an actual human language is, however, a large question. We’ll touch on some possible problems below, and elsewhere I discuss some more.³⁹

Now, as is apparent from my earlier example, Davidsonian contexts may include intentionally characterized features: recall the feature what the speaker referred to in uttering ‘it’ invoked in providing a T-sentence for ‘It’s green.’⁴⁰ (So, again, this conception of context is not available if one is pursuing certain reductionist strategies.) Two worries might arise here. First, it may be worried that the threat of triviality looms. While there’s no reason to think that the contextual features adverted to will include what proposition is expressed, we see that other arguably semantic phenomena will be invoked, as reference just was in our example. The worry, note, doesn’t so much concern the invocation of intentionality per se, as the invocation of a certain sort of intentionality, specifically a kind of linguistic phenomenon. Once again, the proper response is to clarify what questions Davidsonian context is supposed to help answer. It is certainly interesting to ask, for instance, what determines a speaker’s reference on a particular occasion of utterance of a context-sensitive referring expression (and what enables hearers to grasp the reference). But that the Davidsonian approach doesn’t answer that question won’t constitute an objection if it doesn’t aspire to answering it. Of course, there would be room for objecting if it begs off answering too many questions (or doesn’t take on enough), or if it separates questions that seem to go together explanatorily. But this doesn’t seem to be the case. In fact, the Davidsonian approach effects a natural division between our standing knowledge of language and the occasional knowledge we must in addition bring to bear in order to understand particular utterances. The truth-theory aims only to capture the former (which does, however, involve specifying what role one’s occasional knowledge—whatever it is—plays on particular occasions). Tackling the latter can then be reasonably left for pragmatics to handle, to the extent that it can.

This division of labor also provides an answer to the second worry. Above I said that Davidsonian context can be understood as aiming for a middle ground between the non-intentional Lewisian locations and the profligately intentional Stalnakerian presupposition sets. The second worry, then, is whether such a middle ground can in fact be found—whether, in particular, Davidsonian contexts can avoid entanglement in the holism of the mental. Now we see how this indeed is possible. For Davidsonian contexts need advert only to those intentionally characterized features that need be directly plugged into the theory’s variables. Their values, in other words, can be simply taken as given, as far as the truth-theory is concerned: what fixes those values in the first place becomes irrelevant.

Although not all of the speaker’s presuppositions need be invoked on this conception of context, still there might remain the worry that the number of intentionally—and indeed non-intentionally—characterized contextual features required might prove quite large. Let’s turn, then, to desideratum 4—the finiteness constraint.

I begin by recalling some well-known pre-history. Lewis, in his early paper “General Semantics,”⁴¹ specified eight contextual features (“a list,” he later noted, “that was long for its day”⁴²) and, wondering whether more were needed, discussed others one might add.⁴³ Cresswell replied to the “agonies of conscience” he saw Lewis enduring “in trying to decide whether [he had] taken account of enough,” by suggesting, with the support of one example, that “it is impossible to lay down in advance what sort of thing is going to count as a context of use… there is no way of specifying a finite list of contextual coordinates.”⁴⁴ Lewis accepted the point:

Contexts have countless features… the dependence of truth on context [is] surprisingly multifarious… Cresswell rightly complained… [there are] indefinitely many features of context.⁴⁵

And it is not uncommon for those who agree simply to refer back to this exchange.⁴⁶ But not everyone did agree. Lycan, for example, replied:

If English speakers and hearers in fact have all the context-interpretive skills requisite to negotiating all the discoverable factors, then (however many and various these factors may be) there is some definite [sc. finite] number of them, even if as theorists we cannot say in advance what this number is. Therefore, we can be confident that the addition of members to our “indices” would come to an end at some point….⁴⁷

And perhaps Lewis conceded this point as well in writing that “[w]e could wait for the end of linguistic inquiry, and define our indices then”—though he added that “the less patient of us may prefer another solution [viz., adverting to his locations].”⁴⁸—Here ends my historical narrative.

Now, while I won’t go so far as to defend Lycan’s argument,⁴⁹ I do want to defend his conclusion—or at least to suggest that no good argument has been given against it. The first point to note is just an application of the point I stressed above. Many have been overly impressed by the multifariousness of contextual features on which—to use Lewis’ language—“truth sometimes depends,”⁵⁰ features that are “relevant” to the determination of extension.⁵¹ But this language is a bit unclear. Yes, just about anything could be relevant—if what is meant, for instance, is that it could play some causal or evidentiary role either in the determination of content or in a hearer’s coming to grasp what was said. But it doesn’t follow that an interesting semantic theory must account for all aspects of these processes. All the Davidsonian need do is isolate those features of context that provide the truth-theory the contextual information it needs—again, however these features are themselves in turn determined. The semanticist, that is, needs to identify those features of context that, as it were, sum up the rest of the contextual situation in a manner sufficient to yield truth-conditions, given the semantic rules embodied in the truth-theory. In sum, from the infinity of possibly “relevant” contextual features, it doesn’t follow that Davidsonian contexts must be infinite.

Still, one may attempt to argue, by appeal to cases, that even Davidsonian contexts must advert to an infinite number of features. But note that if a language’s lexicon and syntax is finite, then the infinity would have to arise already from a single term or rule.⁵² Is there reason to think the axiom for any single lexical item, for instance, would require an infinity of contextual features? Predications of ‘green’ might be contextually sensitive to the relevant part of the subject of predication, the color’s etiology, facts about the viewing conditions, etc.⁵³ But could this list go on forever? Some argument would be needed to give us reason for thinking so. There is much complexity, to be sure (surprisingly, perhaps—though in retrospect maybe this is what we should have expected), but from complexity doesn’t follow infinity. It is sometimes suggested, however, that there is an open-endedness to our language use, in that as new situations arise, so will new ways of deploying our existing words: mountains are high, but so are tones—and there’s no end of fields to which the high/low contrast could come to be applied. Similarly, there may be no end to the kinds of contextual feature that could become relevant to assessing predications of ‘green.’⁵⁴ The proper response to this, however, is that semantics (or at least this chapter of it) need only be in the business of describing how our language is now. It needn’t as well account for all the ways it might expand over time.⁵⁵ The worry might remain that, although the number of contextual features admitted on the Davidsonian conception has not been shown to be infinite, it has been shown to be quite large—and therefore perhaps unmanageable. But here the Davidsonian may reasonably shift the burden back: if this is to become a full-fledged objection, the demand for manageability must both be clarified and motivated. After all, as mentioned, at this late date we perhaps should not be surprised by the complexity of our linguistic competence.

I turn, finally, to desideratum 6—that contexts be context-insensitively characterized. If among the posited contextual features are propositional attitudes, then context-insensitively characterizing particular contexts will face the same obstacles encountered by Stalnakerian presupposition sets. In addition, the values of many non-intentional features would seem to resist context-insensitive characterization. Consider the sentence ‘That style of play is risky.’ A relevant contextual feature would seem to be the contextually relevant style of play. But unlike with times and places, we seem to lack a general context-insensitive way to identify arbitrary styles of play. Though we may be able to draw attention to aspects of the particular style, we may lack the resources to single it out in any way other than demonstratively.⁵⁶ And so we wouldn’t be able to context-insensitively refer to an arbitrary context. But perhaps this needn’t bother us, so long as we can context-insensitively refer to those contexts to which we have some reason to refer. And, in any case, for all that’s been said, perhaps we might have no reason not to allow ourselves to refer to contexts using context-sensitive terms.

More interesting is the question whether we can characterize what contexts are context-insensitively. For if we can’t, then we are blocked from context-insensitively stating the truth-conditions of the sentences of our target language. The T-sentences for context-sensitive sentences, recall, have the following form:

(u)(x)(y)…(z)[If uR(S, x, y,…, z), then {u is true iff F(x, y,…, z)}].

For example, the T-sentence capturing the variable reference of ‘it’ in ‘It’s green’ would be:

(u)(x)If u is an utterance of ‘It’s green’ in which ‘it’ refers to x, then (u is true iff x is green).

The T-sentence thus adverts to the contextual feature what ‘it’ refers to. If this contextual feature is itself context-sensitively characterized, then the T-sentence as a whole will be as well. But if the T-sentence is context-sensitive, it does not—considered independently of any particular context of utterance—express a proposition at all; and so, in particular, it doesn’t specify the sentence’s truth-conditions. The Davidsonian semanticist must therefore either allow that the content of the truth-theory itself can vary across contexts of utterance, or she must ensure that she uses only context-insensitive terms in characterizing what the relevant contextual features are. If ‘refers’—or any other expression used to characterize a contextual feature—is subject itself to context-sensitivity (even if only through being subject to vagueness), then taking the latter tack will require stipulating this context-sensitivity away. Note that this requires that the individuation conditions of the objects quantified over be fixed as well.⁵⁷

4 Conclusion

This completes my examination of these three conceptions of context. We have seen that each satisfies some of the six desiderata, while violating others. And in each case, we have seen as well how these violations needn’t constitute an objection, so long as one is clear about what projects each conception may and may not be suited to subserve. Not clearly delineating these projects leads only to confusion. It is my hope that this examination not only helps sort out what is—or ought to be—at issue in discussions of context, but also sheds some light on the nature of these varying projects themselves.

I close by addressing one reason why some of the issues I have tried to separate are sometimes lumped together—viz., a not unreasonable desire to treat language and linguistic phenomena uniformly. It would seem a laudable goal to seek an explanatory framework that could accommodate all of the projects we have discussed. The desire for such motivates searching for a single conception of context suitable for all our linguistic needs. I have not argued that there could not be such a thing—beyond, that is, gesturing towards the possible intractability of certain areas.⁵⁸ But note that even if there were a single conception of context suitable in principle for all projects aimed at explaining linguistic behavior, it would not follow that adverting to such contexts would yield the most perspicuous explanation no matter the issue. For example, employing Stalnakerian contexts, and the highly complex rules they would require, would certainly shed less light on specifically what we know in knowing a particular language than would the use of Davidsonian contexts. (I’m supposing here that both approaches could pan out in principle, and that Stalnaker’s would thus be suitable to a greater range of questions than the Davidsonian’s.) Various specifically semantic facts, explicitly highlighted on the Davidsonian approach, would get lost in the complex semantic/pragmatic brew the Stalnakerian approach would yield. We would thus still have use for more than one conception of context.⁵⁹