INTRODUCTION
The linguistic variable was introduced in the 1960s by William Labov as a methodological tool with which to bring together and co-examine surface forms that alternate in discourse. In proposing this, Labov was reacting to an earlier view of the units of linguistic analysis as “1. discrete, 2. invariant, 3. conjunctively defined, and 4. qualitatively different” (1966:19), which he termed the “categorical view” (ibid.). In contrast, Labov claimed, “the speech performance of the individual or the group is best explained through the assumption of an underlying linguistic continuum, in which categories form, reform, and dissolve” (1966:21). The driving force behind the formation of linguistic categories is some conditioning factor that regulates the distribution of surface variants: “if the conditioning factor disappears completely, the linguistic variable loses its ordered structure, and is resolved into one or more phonemic categories” (1966:16). This conditioning factor may be intralinguistic (other linguistic variables) but I would argue that the most interesting cases for the sociolinguist are those in which the conditioning factor is extralinguistic (socio-economic class, ethnicity, age, etc.),Footnote 1 since this opens a window on the evaluative norms of a community of speakers and how linguistic variants are invested with social significance.Footnote 2
This is a tall order—both because the precise identification of variants of the same variable, as well as their significance, may be in constant flux—and, to fulfill it with any degree of methodological rigor, the class of variants claimed to constitute a linguistic variable should, according to Labov (Reference Labov1966, Reference Labov1972), be:
- (1)
a. ordered along a continuous dimension;
b. highly frequent;
c. immune to total suppression;
d. codable;
e. widely distributed throughout the population; and
f. their position should be determined by an independent linguistic or extralinguistic variable.
This conception of the linguistic variable has provided tremendous impetus to the study of variation over the past forty years leading to significant findings, especially in socio-phonetics. At the same time, it has proved a limiting influence on the study of variation at other levels, especially in syntax and discourse. As Watt (Reference Watt, Llamas, Stockwell and Mullany2007:3) admits, “[t]he study of discourse variation is still at an early stage, and while it presents challenging problems … the fact that such variation has been found to be systematic indicates that a full understanding of how speakers construct conversations will necessitate a good deal of further research to establish more explicitly the forms, functions, and uses of discourse variables.”
This article revisits the thorny question of the linguistic variable's applicability beyond phonology, where it has proved most fruitful, to pragmatic variation in particular. Raised in the 1970s, this question was hotly debated by researchers until the early 1990s, when it was set aside with inconclusive results, only to be resurrected—albeit implicitly—by ongoing work in variational pragmatics (Schneider & Barron Reference Schneider, Barron, Schneider and Barron2008). Clearly, the success of the variational pragmatic enterprise is closely intertwined with the usefulness of the variationist framework for analyzing pragmatic variation, which in turn hinges on providing a principled definition of the pragmatic variable.
To explore this possibility, I begin by elucidating the nature of the Labovian variable as applied to phonology and trace current developments in thinking about phonological variables. I then turn to the debate about applying the notion of the linguistic variable beyond (morpho)phonology and review both the difficulties arising in this respect, as well as how researchers have proposed to deal with them. These discussions lead to an updated view of the linguistic variable based on the notion of functional equivalence. For those interested in pragmatic variation, the question then becomes how to define functional equivalence specifically in pragmatics. Theoretical advances in pragmatics, in particular the distinction between conceptual and procedural meaning elaborated within Relevance Theory (Sperber & Wilson Reference Sperber and Wilson1986/1995:258–59; Blakemore Reference Blakemore2002), are promising in this respect. In the final section, I propose the notion of procedural meaning as the theoretical tool that was missing twenty years ago and that can supply a unifying basis for defining pragmatic variables, and I demonstrate the applicability of this proposal to three recent analyses of pragmatic variation in the literature. This proposal is in line with Labov's original conception of the linguistic variable “as a concept, as a working tool, and as a structural element of analysis” (Labov Reference Labov1966:19), “an abstraction” (1966:21), a “heuristic device” (Labov Reference Labov1978:1), all of which suggest that he did not envisage linguistic variables as psychologically real to speakers themselves.Footnote 3 A desirable consequence of decoupling the linguistic variable from the question of psychological plausibility is that it allows for more abstract, theoretically bound interpretations of this notion, which are also framework-specific (as e.g. proposed by Cornips Reference Cornips and Brown2005 for syntax; see also Extending the linguistic variable: Functional equivalence). In other words, more than simply empirically adequate to describe the types of pragmatic variation that happen to have been studied, procedural meaning as expounded within Relevance Theory is precisely the kind of abstract theoretical construct one might expect to find at the heart of the pragmatic variable.
ELUCIDATING THE LABOVIAN VARIABLE
I listed earlier six requirements that surface forms must meet to be considered variants of the same variable (1a–f). One thing that is frequently overlooked is that these six requirements are not all equally important: some are constitutive of what makes surface forms variants of the same variable and so required of every linguistic variable, while others are merely desirable and hence optional. Specifically, requirements (1a), that variants be ordered along a continuous dimension, and (1f), that their position on this dimension be determined by an independent linguistic or extralinguistic variable, are intrinsic to the variationist enterprise and therefore sine qua nons of applying this methodology to the data. Labov spells out their importance as follows:
The empirically established statement of the continuously divisible character of the linguistic variable is a necessary condition for its use in structural analysis. Variation must be (1) ordered, and (2) correlated with some independent variable. If such strict conditions are not laid down, the linguistic variable would provide an easy solution to almost any structural problem, but at the cost of eliminating all rigorous method. (1966:21)
The idea, at its most general, is that there should be a linguistic dimension capturing the similarity among variants of a linguistic variable (justifying their inclusion under the same variable), while any differences in their use should be possible to motivate with reference to another, independent variable, be that linguistic or extralinguistic. However, that does not mean that there is only a finite number of ways in which variants may be alike—in other words, the possible dimensions of variability are not a closed set. As Dines emphasized, “if seemingly unlike elements can be considered alike according to some criterion, then this criterion becomes the basis of the variable—a simple classificatory procedure. There will thus be many potential dimensions or levels of linguistic structure where variables may be grounded and what these are may need to be determined anew with every study” (1980:15). This functional interpretation of the linguistic variable is already present in Labov's thinking from the start. As the following comment suggests, what he had in mind was not some system-external, objective similarity between variants but rather a more malleable notion, what we might call an emergent similarity created in response to some (external) conditioning factor: “if the conditioning factor disappears completely, the linguistic variable loses its ordered structure, and is resolved into one or more phonemic categories” (Labov Reference Labov1966:16). Linguistic variables, then, are formed in response to linguistic or extralinguistic conditioning factors—that is, they are dynamically emergent.
The emergent character of the linguistic variable is further supported by Labov's examples of phonological variables. Take postvocalic /r/: its two variants are [r] (presence of weak or strong consonantal constriction) vs. Ø (absence of constriction; Labov Reference Labov1972:72–73). The only continuous dimension on which these can be said to be ordered is a trivial one, viz. presence of a phenomenon (constriction) vs. its absence. Despite not attempting a finer subdivision or more detailed ordering of its phonetic realizations, Labov is able to reach a rich set of conclusions about its social and stylistic significance. Arguably, then, rather than a linear ordering of variants along a single continuous dimension—a claim that seems to be problematic even for phonological variants (Hudson Reference Hudson1980/1996:174–75; Romaine Reference Romaine1981:30, n. 4)—what we are dealing with may be more generally construed as a multidimensional space (as e.g. proposed by Le Page & Tabouret-Keller Reference Le Page and Tabouret-Keller1985), in which variants occupy different positions determined by a conditioning factor external to the variable itself. What guarantees “the structural significance of the variable,” in this case, is the fact that “it alters in an orderly way when one or more of the external social variables change” (Winford Reference Winford1984:268, emphasis added).Footnote 4 The important insight is that, to be considered variants of the same variable, surface forms must, at one and the same time, be similar (in some relevant linguistic dimension) and systematically—or, predictably—different (in another, linguistic or extralinguistic, dimension). Similarity ensures that speakers continue to be able to communicate despite the variation (Labov's “sameness of representational function,” 1982:25–26, 53; see also Critiques of the Labovian variable and proposed solutions), while systematic/predictable difference invests the variation with social significance (Labov's “normal/structured heterogeneity,” 1982:17–18). Requirements (1a) and (1f) above capture this foundational premise of the variationist paradigm, making them, more than inter-related, two sides of the same coin.
By contrast, the requirements of high frequency, codability, and wide distribution throughout the population ((1b), (1d), and (1e)) are rather incidental. They emanate more from Labov's empirical focus on phonological variables (which ipso facto meet these requirements; cf. Labov Reference Labov1966:6; Hudson Reference Hudson1980/1996:172–73) rather than theoretical necessity, and respond to the exigencies of statistical analysis, since they offer better guarantees of reaching desired levels of significance.Footnote 5 So much follows from another remark by Labov, listing high frequency and wide distribution across ordered strata of society among “the most useful properties of a linguistic variable to serve as the focus for the study of a speech community” (1972:7–8, emphasis added), suggesting that these requirements are neither necessary nor sufficient for successfully identifying a linguistic variable in attested data.
Finally, the linguistic variable's alleged immunity to total suppression (Labov Reference Labov1966:6, Reference Labov1972:9; requirement (1c)), has generated some discussion lately. Labov's original inclusion of this requirement seems to have been motivated by two related observations: synchronically, stigmatized forms persist, however attenuated, into the most careful speech styles of their users (1972:103–5), while diachronically, they resist prescriptivist efforts to reverse change under way (1982:29). The latter observation, however, does not mean that reversals never occur (1982:59); rather, awareness seems to operate at two different levels. Variation in low-level output rules (such as might lead to regular sound change, Labov Reference Labov, Lehmann and Malkiel1982:62) and/or during the early stages of change (Labov Reference Labov, Lehmann and Malkiel1982:31) is hardest to suppress because it falls “below the level of social awareness. No-one in the community refers to the change and it is hard to bring it to consciousness” (1982:80). As the social significance of variation increases, so does awareness of it, such that “late stages of change are often well-known stereotypes” (1982:31). Consequently, variation involving more abstract rules or processes (that are prone to change by lexical diffusion, 1982:63) and/or during later stages of change may be potentially suppressible. This awareness-based explanation offers an alternative account of why some changes remain incomplete. More importantly, it suggests that, even for Labov, immunity to total suppression was not all-or-nothing: he was careful to qualify it not only by degree (“immunity to total suppression”), but also by the ways it could be implemented (“style shifting is not directly governed by overt values,” 1972:103, emphasis added), leaving open the possibility of a less mechanistic process that may be sensitive to speakers' implicit biases—which is precisely what some of the latest research has been finding.Footnote 6
Comparing the vowel production of New Zealand subjects before and after listening to an Australian talker (cross-dialect variation), and of White California subjects before and after listening to a Black talker (within-dialect variation), Babel (Reference Babel2009) found that phonetic convergence—though not maintenance or divergence—positively correlates with social factors such as implicit biases toward groups and attractiveness, as well as the vowel's social significance synchronically within the dialect. Based on these findings, she concluded that spontaneous imitation is automatically influenced by social factors in the shape of implicit socio-cultural biases. This discovery sheds new light on the claimed immunity of phonological variants from total suppression. While not subject to conscious manipulation, speakers' use of phonological variants already incorporates their implicit social biases, making phonological variation the outcome of an intricate interplay of factors in which “social evaluation and attitudes play [more than—MT] a minor role” (Labov Reference Labov2001:20, quoted in Trudgill Reference Trudgill2008:252). Since much of language use similarly obeys a principle of least effort (Zipf Reference Zipf1949),Footnote 7 this finding closes the qualitative gap between phonological and other types of variation, all of which seem to be subject to socially constituted automaticity to some extent.
On closer inspection, then, of the six requirements outlined in (1), only two constitute necessary and sufficient conditions for surface forms to be considered variants of the same variable. These are that variants be (i) ordered (i.e. cogently placeable in a linguistic space that expresses their similarity), and (ii) correlated with some independent variable. These two requirements capture the premises of sameness of representational function and normal/structured heterogeneity, which lie at the heart of Labov's insight that variation in language is not random but itself a carrier of information, in this case social information. In the next section, I investigate how these two requirements have guided the implementation of the linguistic variable to non-phonological variation, and discuss both old and new problems arising in this respect, before surveying proposed solutions to these difficulties.
CRITIQUES OF THE LABOVIAN VARIABLE AND PROPOSED SOLUTIONS
“Two ways of saying the same thing”
In Labov's work, the requirement that variants of a linguistic variable should be “alternate ways of saying ‘the same’ thing” (Labov Reference Labov1972:188) springs from two interrelated concerns. The first is a diachronic one, specifically “how can meaningful elements be variably deleted without affecting the ability of speakers to communicate representational information?” (Labov Reference Labov, Lehmann and Malkiel1982:53). The second concern is synchronic, namely ensuring that the reason for the alternation between variants does not inhere in the variants themselves but lies beyond them, in other linguistic or—more pertinently for present purposes—social factors (Lavandera Reference Lavandera1978:173). To address these two concerns, Labov proposes that variants must be equivalent in meaning, which thus emerges as a precondition for variationist analysis. The problem is that there are many kinds of meaning—and, depending on which kind one prioritizes, one is bound to define meaning equivalence differently, producing different delimitations of the linguistic variable each time.
For Labov, the emphasis—as also revealed by his phrasing of the first, diachronic, concern above—ought to be placed on “the representational function of language” (related to the world), that is, following Bühler (Reference Bühler1934/1990), “the capacity to represent immediate, distant, imaginary and even impossible configurations of our perceptual and conceptual experience” (Labov Reference Labov, Lehmann and Malkiel1982:25–26). This representational function is distinct from the “expressive” (related to the speaker) and “appeal” (related to the hearer) functions, which Labov considers to be subordinate to it (ibid.; cf. 1978:2). The answer, then, to the puzzle of language variation and change is representational sameness: “[t]he alternate ways of saying ‘the same thing’ which form the heterogeneous character of the community are representational sames, equivalent in this sense of meaning” (Labov Reference Labov, Lehmann and Malkiel1982:25–26). How, then, to define representational sameness? Despite acknowledging that “any narrow use of truth-conditional semantics is bound to fail, and a number of extensions are needed to make use of that notion for natural language” (Labov Reference Labov, Lehmann and Malkiel1982:85, n. 2), Labov's answer to this second question is, in the end, no less categorical: “I would like to say that two utterances that refer to the same state of affairs have the same truth-value and … limit the use of ‘meaning’ to this sense” (1978:2; cf. 1972:188, n. 7; Weiner & Labov Reference Weiner and Labov1983:30–31). In this way, in Labov's account, meaning is effectively ‘neutralized’: whether dealing with phonological variants (which express no meaning in and of themselves but only as part of word forms considered surface realizations of the same underlying lexeme; cf. Romaine Reference Romaine1981:9, Winford Reference Winford1984:269–70) or with syntactic ones (which should “have the same meaning in a truth-conditional sense [in that] they are used interchangeably to refer to the same states of affairs,” Weiner & Labov Reference Weiner and Labov1983:30, emphasis added), the alternation between variants of a linguistic variable does not affect (truth-conditional) meaning, which can thus be safely ignored.Footnote 8
The assumption of truth-conditional equivalence is, however, not unproblematic. To begin with, truth-conditional equivalence furnishes the basis for delimiting variable contexts, that is, those linguistic environments where variants of a linguistic variable can alternate without affecting the truth conditions of the utterance—which may well be only a subset of the possible contexts of occurrence for each variant (Weiner & Labov Reference Weiner and Labov1983:31; cf. Tagliamonte Reference Tagliamonte2006:86–94). The problem, already highlighted before, is that the explanation for the alternation of the variants in the variable contexts often lies beyond them, and in those contexts in which the variants cannot be used interchangeably. For instance, as Lavandera writes:
in the very simple case of ‘wiped out’ vs. ‘exhausted’ if we consider only those cases where ‘wiped out’ communicated referentially the same thing as ‘exhausted’ (i.e. where they can be used to describe the same state of affairs) we are perhaps not going to be able to explain why ‘wiped out’ is the more colloquial form of the two. … only if we consider those cases where ‘wiped out’ means or communicates ‘demolished’, can we then view it as a metaphor and proceed to further explanation. (1978:179)
Similarly, considering the alternating negative preterite constructions I never went to school today vs. I didn't go to school in British English, Cheshire cautions:
if we consider this use of never simply in terms of its alternation with didn't, we overlook a more general aspect of negation in English. … the variation that occurs between the negative preterite forms (never + simple past tense verb and didn't + uninflected verb) is part of the system of negation in English, in which never is the emphatic, marked negator and not is the neutral, unmarked negator. Analyzing this variation in terms of a simple alternation in the forms of the negative preterite, therefore, overlooks the way in which these forms fit into the system of negation in English. (1987:269–70)
In other words, limiting analysis to only the variable contexts leaves us unable to go beyond description and explain the motivation for the variation. This criticism is particularly poignant in light of Labov's construing the linguistic variable as a “structural” unit, “integrated into a larger system of functioning units” (1972:8), suggesting instead that, to achieve precisely this aim, the entire distribution of the alternating variants should be studied.Footnote 9
Studying the structural integration of variants may furthermore be needed not only for explanation but also for description. As Winford points out regarding variationist studies of creole continua:
what is involved is alternation between forms which operate in the creole system, and their ‘equivalents’ in Standard English as well as in the ‘intermediate’ varieties of speech which form the continuum between the creole and the English system. … A closer examination, however, reveals that the bases of equivalence vary from one case to another …. each of these items has a place within the respective basilectal, mesolectal and acrolectal verb systems which are clearly distinct in the organization of their grammatical categories. … It follows that decisions concerning the semantic content and functions of such forms rest partly on analysis of the respective systems within which they function. Any a priori assumption of equivalence between such forms would therefore seem to beg the question. (1984:275–76)
Like Lavandera (Reference Lavandera1978) and Cheshire (Reference Cheshire1987), Winford's point is that assessing equivalence solely on the grounds of truth conditions fails to do justice to other dimensions of the meaning of variants emanating from their participation in larger systems of grammatical and semantic signification.Footnote 10 Grammatical categories cannot be simply presumed to be isomorphic across varieties; rather, they must be empirically shown to be, each time anew.
This point is reminiscent of developments in semantics/pragmatics, where truth-evaluable propositions are increasingly viewed as the output, not of linguistic semantics alone, but of decoding and pragmatic inference operating in tandem (Sperber & Wilson Reference Sperber and Wilson1986/1995; Levinson Reference Levinson2000:165ff.), prompting the formulation of different versions of contextualism and the adoption of new labels such as truth-conditional pragmatics (Récanati Reference Récanati2004, Reference Récanati2010). On this view,
the meaning of words is adjusted or ‘modulated’ so as to fit what is being talked about. Sense modulation is essential to speech because we use a (more or less) fixed stock of lexemes to talk about an indefinite variety of things, situations and experiences. Through the interaction between the context-independent meanings of our words and the particulars of the situation talked about, contextualized, modulated senses emerge, appropriate to the situation at hand. (Récanati Reference Récanati2004:189)
Moreover, “modulation is truth conditionally relevant, because the words' contribution to truth-conditional content … is not their pre-modulation meaning but their modulated senses” (Récanati Reference Récanati2004:193, original emphasis). It follows that establishing the truth-conditional equivalence of variants must take into account not only linguistic context but also the real-world situation surrounding the act of utterance. This invalidates attempts to substantiate the truth-conditional equivalence of linguistic variants based on linguistic context alone. In other words, the whole exercise of delimiting variable contexts on the basis of truth-conditional equivalence is excluded. Indeed, the often stipulative flavor of the results of this exercise has not gone by without criticism (cf. Garcia's 1985:214–15 discussion of Dines's (1980) results).
Another problem for truth-conditional equivalence is that it relies on the existence of a clear line demarcating truth-conditional from non-truth-conditional meaning. The existence of such a line, however, is doubtful in light of, not only such classical controversial notions as conventional implicatures (which are indefeasible but do not enter into truth-conditions) and generalized conversational implicatures (which are defeasible but do; Grice Reference Grice, Cole and Morgan1975/1989), but also of what one might call cultural loanwords, that is, expressions whose non-truth-conditional meaning carries over into, and colors, their truth-conditional one. Consider (2):
(2) Be it Tapas, Mezedes, Dim Sum or Appetizers, this is my favourite way to eat. Tasty platter you've got there. [http://theculinarychase.blogspot.com/2009/05/tapas.html; accessed May 27, 2009)
Posted under a recipe for fingerfood, tapas, mezedes, dim sum and appetizers are used by the author of (2) to refer to the same real-world situation, and so are treated as referentially equivalent. This is also suggested by their interchangeability in the collocations Greek tapas (3,340 Google hits), Chinese tapas (2,030 hits), Spanish dim sum (467 hits), and Greek dim sum (134 hits), which is similarly predicated on the referential dimension of their meaning. However, this similarity does not exhaust their meaning, which is why the author of (2) does not feel it is redundant to juxtapose them as s/he does. The more general point that examples such as (2) help drive home is that truth-conditional equivalence—granting that it is possible to establish—does not guarantee that two expressions represent “two ways of saying the same thing”: inherent differences in non-truth-conditional/pragmatic meaning may still exist and it is these differences, rather than an independent linguistic or social factor, that may motivate the choice between them; in other words, the two forms, despite being truth-conditionally equivalent, may ultimately express different messages (Lavandera Reference Lavandera, Dittmar and Schlieben-Lange1982:88–89).
The way out of this epistemological impasse is not to define meaning equivalence ever more strictly as encompassing both truth-conditional and non-truth-conditional equivalence, for such a move would make the task of the variationist an (almost?) impossible one. Rather, it may be time to consider seriously the possibility that not only forms but also meanings—what Lavandera (Reference Lavandera, Dittmar and Schlieben-Lange1982) has called social signifieds—may be differentially distributed across groups (Dines Reference Dines1980:15, 25; Garcia Reference Garcia1985:210, 220; Eckert Reference Eckert2008:456, 467). This is a possibility that ethical considerations seem to have placed out of bounds for researchers during the past several decades. Nevertheless, at a time when consensus views of society are being increasingly criticized as simplistic and the ideal of equality is becoming more and more decoupled from similarity, as sociolinguists it may be time for us too to stop focusing exclusively on those referential meanings that are shared across groups, shying away from those that are not.
As mentioned at the start of this section, the idea that linguistic variants must be truth-conditionally equivalent follows from an emphasis on “the primacy of the representational function in the organization of human language” (Labov Reference Labov, Lehmann and Malkiel1982:25). Since the representational function concerns the relationship between the sign and the world, it was natural to interpret representational equivalence as equivalence in the states of affairs denoted by two utterances, that is, as truth-conditional equivalence. However, Bühler, who proposed the original trifurcation into representational, expressive, and appeal functions, did not assert the primacy of one over the other; rather, he believed that “linguistic meaning is not exhausted by any one of these three relations” (Habermas Reference Habermas and Cook1998:278). This interpretation frees language from the requirement that it be defined, above all, by a relation of isomorphism with the world, opening the door to other—performative—possibilities.
Scholars have been aware of these possibilities for some time. For instance, Romaine noted that “reference is but one of the many functions of utterances and not the basis of all others” and highlighted the work of scholars such as “Firth (Reference Firth1966) [who] treated the descriptive function of language as something that was subsidiary to and part of the more important and general social (i.e. indexical) function of language” (1981:21). A recent suggestion along these lines gives center-stage to the notion of intersubjectivity that treats linguistic expressions as cues to cognitive coordination between speaker and addressee:
engaging in cognitive coordination comes down to, for the speaker/writer, an attempt to influence someone else's thoughts, attitudes or even immediate behavior. For the addressee it involves finding out what kind of influence it is that the speaker/writer is trying to exert … The default condition for ordinary expressions is that they provide an argument for some conclusion, and this argumentative orientation is what is constant in the function of the expression, while its information value is more variable. (Verhagen Reference Verhagen2005:10, original emphasis)
Contrary to the representational view advocated by Labov, on the intersubjectivity view, the informative function of language is subordinate to the argumentative one (cf. Anscombre & Ducrot Reference Anscombre and Ducrot1983:169), allowing a hitherto neglected aspect in variationist analyses, that of speaker agency, to come to the fore and provide an alternative basis for the analysis of variation. As Eckert remarks, “speakers' agency in the use of variables has been viewed as limited to making claims about their place in social space by either emphasizing or downplaying their category membership through the quantitative manipulation of markers” (2008:455). However, such a view yields a rather static picture of the social meaning of variation, offering no insight into how the social significance of the same variant may be impacted by its use in different contexts, what Eckert has called “an indexical field, or constellation of ideologically related meanings, any one of which can be activated in the situated use of the variable” (2008:454, original emphasis).
Shifting our focus from the relation of language to the world to the relation between speaker and addressee allows not only a renewed appreciation of the role of speaker agency in language variation but also some insight into what remained, under the representational view, a puzzle—acknowledged by Labov himself—namely, that not all change proceeds according to the principle of preservation of representational (read: truth-conditional) meaning; instead, changes can also be disruptive of communication. Recently, the fact that “sound change has led to a considerable degree of mutual unintelligibility of the phonologies of North American dialects” (Labov Reference Labov2001:12) has been cleverly capitalized on in a TV commercial for Carfax, a service providing used-vehicle history reports to help consumers make informed buying choices.Footnote 11 In the commercial, a young couple is in a dealer's office after having just test-driven a car:
(3) Husband: The car is perfect. Show me the [karfaks].
Dealer: Sure. Here you go! ((produces a hand puppet that looks like a fox and is wearing a T-shirt with CAR FOX written on it))
Husband: What's that?
Dealer: It's a [karfaks]! ((points to the letters on the T-shirt))
Wife: He said [karfæks]!
Dealer: ((in a shrill)) Hi! I'm your [karfaks]! Buy the car, you're gonna love it! ((in his normal voice)) [karfaks] ((in a shrill) [karfaks] ((in his normal voice)) same thing, uh?
The fronting of /α/ to [a]—a fragment of the Northern Cities vowel shift—captured in this commercial clearly affects truth-conditions (what is taken to be the referent of [karfaks] each time), giving rise to the humorous effect. Examples such as (3) illustrate that preservation of truth-conditional meaning is not a (major) consideration in language change, ultimately leading Labov to view “linguistic heterogeneity and change … as non-functional and, in some sense, ‘pathogenic’” (Verleyen Reference Verleyen2006:350, my translation). An alternative explanation begins to emerge if language is seen, not as an instrument for communicating accurately about the world, but rather as a tool for influencing one's conversational partners and making them “see the world through one's eyes.” Placing intersubjectivity, rather than exchanging information about the world, at the basis of our account of language variation and change will enable us to view changes as emanating from the social indexicality of linguistic forms, which speakers draw on according to their dispositionsFootnote 12 as they strive for cognitive coordination with their addressees.
Extending the linguistic variable: Functional equivalence
In their quest for an alternative basis on which to define the linguistic variable given the inadequacy of truth-conditional equivalence in this regard, researchers over the past thirty years seem to have reached a consensus in the notion of functional equivalence. Lavandera (Reference Lavandera1978:181) proposed “to relax the condition that the referential meaning must be the same for all the alternants and substitute for it a condition of functional comparability,” which may be further spelled out in terms of a common grammatical function (ibid., n. 7). In a similar vein, Dines proposed “that variables may be postulated on the basis of common function in discourse [which] does not preclude the possibility of a semantic tie existing between variants, but it removes the necessity of one” (1980:15–16, original emphasis). Summarizing this approach, Tagliamonte suggests that “the linguistic variable is an abstraction. The varying forms must exist in some linguistically meaningful subsystem of the grammar” (2006:76) and be characterized by the properties of “1) functional equivalence; 2) distribution, and 3) structural embedding” (2006:97). Functional equivalence is, however, no longer spelled out as semantic equivalence but rather as “weak complementarity across the speech community” (2006:76), whereby “the proposed variants can serve one, or more generally, similar discourse functions” (Sankoff & Thibault Reference Sankoff, Thibault, Johns and Strong1981:208 quoted in Tagliamonte Reference Tagliamonte2006:75, emphasis added).
A more radical interpretation of functional equivalence is proposed by Cheshire, for whom “precedence has to be given to the semantic element that appears to be constant, though the grammatical category [may be] indistinct” (1987:277). Crucially, this semantic element should not be understood in truth-conditional but rather in cognitive terms: “it seems likely that other types of linguistic variation are governed by cognitive processes also, and this may well be a fruitful avenue to explore” (1987:271). In recent work, Cheshire develops this idea further:
The variable may well be a heuristic construct that does not necessarily map onto the units of linguistic structure. (2005:494)
Thus in order to see social variation involving syntactic forms it is helpful to adopt a broad-brush approach that focuses on the choices speakers make from all components of their knowledge of language, not just the syntactic component. (2005:480)
[A] full analysis of the forms fulfilling the same discourse function … shows speakers drawing on a range of linguistic features. They include syntactic forms made available by the grammar, discourse markers and pragmatic particles associated with discourse management and the construction of interpersonal relations as well as performance features stemming from the demands of online production. (2005:489)
The idea that the locus of variation may lie at the interface of the syntactic with other components of linguistic competence is also found in studies of syntactic (micro)variation from a generativist perspective. In this context, Cornips & Corrigan suggest that:
the interface between the syntax module and the domains of discourse/pragmatics (i.e. the interaction module where ‘soft’ constraints are violated) is the arena in which variation will be most frequent or emerge more easily than would be the case in the syntax module alone. From this, we may put forward the hypothesis that variation that has social meaning is more frequently located in this domain than it is in others. (Reference Cornips, Corrigan, Cornips and Corrigan2005:14)
They furthermore speculate that this domain “may well be the left periphery which is linked to the specification of the illocutionary force of the utterance … and ultimately to the speaker's pragmatically-motivated choice” (2005:22, n. 12). Identifying linguistic variables with the abstract features of grammatical formalisms, Cornips & Corrigan further argue that “variationist sociolinguists who resort to formal linguistic theory can find novel and more effective measures for deciding which variants are syntactically related and which are syntactically remote” (2005:8). A consequence of this approach is that “the syntactic variable as the structural unit have [sic] to be defined again depending on the particular model of syntax one employs. … Since the consecutive models in Generative Grammar follow each other at rather high speed different make-ups of the structural unit can be detected in the last decades” (Cornips Reference Cornips and Brown2005:31).
Scrutinizing the interpretations of functional equivalence in the studies cited above, one may discern two possible lines of interpretation that parallel the ways in which the “why” question—as in Labov's “Why does anyone say anything?” (1972:207), in response to which Lavandera proposes functional equivalence in the first place—can be answered. Interpreting the question “why” as asking “To what end?”—as Lavandera does—leads us to a teleological interpretation, whereby linguistic variants are considered equivalent if they can be used interchangeably in order to achieve similar perlocutionary effects in discourse. However, this teleological interpretation would be categorically rejected by at least some functionalists, including Martinet (Verleyen Reference Verleyen2006:339, n. 10). Alternatively, the “why” question may be interpreted causally as asking “On what grounds?” On this second view, linguistic variants are equivalent if they represent alternating realizations of a single abstract linguistic category. In other words, the studies cited above seem to be using two competing understandings of function, a communicatively oriented one (a variant's function vis-à-vis the speaker's intentions) and a structurally oriented one (a variant's function vis-à-vis other items in the linguistic system), respectively.
Consequently, while the notion of functional equivalence is certainly in the right track in terms of extending the linguistic variable to nonphonological variants, it does not by itself carry all of the answers but must be specified further. The remainder of this article explores how functional equivalence can be operationalized to account for pragmatic variation, putting forth a novel proposal that capitalizes on the notion of procedural meaning (Blakemore Reference Blakemore1987, Reference Blakemore2002; Wilson & Sperber Reference Wilson and Sperber1993). But first, we must define what is meant by pragmatic variation.
VARIATIONAL PRAGMATICS AND THE STUDY OF PRAGMATIC VARIATION
In a recent volume entitled Variational pragmatics, Schneider & Barron propose the discipline of variational pragmatics “conceptualized as the interface of pragmatics with variational linguistics, i.e. with modern dialectology, as a branch of contemporary sociolinguistics” (2008:1). The aim of the new discipline is, broadly put, to problematize the underlying assumption that populations of native speakers are homogeneous with respect to their pragmatic choices. Rather, variation can be found in pragmatics as much as at any other level of analysis. Schneider & Barron distinguish between two kinds of factors that condition pragmatic variation: macrosocial factors, which “remain relatively stable across situations and contexts” and include region, social class, ethnicity, gender, and age; and microsocial factors, which “may change from situation to situation” and include relational variables such as power and distance, and style (Schneider & Barron Reference Schneider, Barron, Schneider and Barron2008:18).
Schneider & Barron view variational pragmatics as explicitly variationist, both in theoretical orientation as well as in the methodology adopted. In terms of the latter, the macrosocial factors they list (2008:16–18, and above) as well as the quantitative approach adopted by the papers in the volume firmly place variational pragmatics within the variationist paradigm. In terms of theoretical orientation too, the authors draw an explicit comparison between the new field and (urban) dialectology (2008:1–3). Moreover, they think of variational pragmatics as “contrastive by definition: variety-specific pragmatic features, including variety-exclusive features as well as characteristic distributions of variety-preferential features, can only be established by comparing two or more varieties of the same language” (2008:21). This remark brings to the fore the notion of tertium comparationis, a notion central to contrastive analysis, referring to the quality that two entities share and which provides the basis for their comparison (Janicki Reference Janicki, Kastovsky and Szwedek1986), clearly implying a conceptualization of these features as variants of a single variable. This variationist conceptualization is further made explicit by the authors through several references to “pragmatic variables” and their “possible variants” (Schneider & Barron Reference Schneider, Barron, Schneider and Barron2008:11, 13, 22). The methodological and theoretical orientation of variational pragmatics thus places at the heart of the new enterprise the question of the pragmatic variable—a question to which, as I have tried to establish in the previous sections, there are no easy answers.
Schneider & Barron list five levels of analysis at which pragmatic variation may be sought (2008:19–21). At the formal level of analysis, one may investigate linguistic forms such as discourse markers, hedges, and upgraders, seeking to establish their communicative functions in discourse. At the actional level, the emphasis is on speech acts and their realization and modification in terms of directness and politeness. At the interactional level, Schneider & Barron propose to study the organization of speech acts into larger sequences of discourse, such as adjacency pairs, and conversational openings and closings, while at the topic level they seek to address the contents of these larger sequences and how these are both selected and managed. Finally, at the organizational level, local sequential patterns involving turn-taking phenomena can be addressed.Footnote 13 Based on this overview of its possible loci, pragmatic variation may be defined as: (i) variation in the range of pragmatic meanings expressed by a single form (form-to-function mappings; Schneider & Barron Reference Schneider, Barron, Schneider and Barron2008:10), and (ii) variation in the range of forms that express a single pragmatic meaning (function-to-form mappings; ibid.). A foundational task of variational pragmatics is, then, to define the pragmatic variable in such a way that it will furnish a single “conceptual” roof for the study of these two types of variation.
TOWARDS A PROCEDURAL INTERPRETATION OF THE PRAGMATIC VARIABLE
Pragmatic variation revisited
It has been claimed that, once we move away from the realm of phonology, variation is less common. For instance, Hudson hypothesized that “differences in syntax tend to be suppressed, whereas those in vocabulary and pronunciation tend to be favored and used as markers of social differences” (Reference Hudson1980/1996:45). In contrast, Romaine argued that “[g]rammatical differences, e.g. multiple negation, have tended to show patterns of sharp stratification … while phonological variables, like the raising of /æ/, reveal fine stratification. … it is difficult to accept Hudson's view that syntax is a marker of cohesion when the syntactic differences which do exist in a community may be so divisive” (1981:5, original emphasis).Footnote 14 An alternative explanation claims that variation in syntax is no less frequent but simply less obvious and, therefore, harder to locate and study (Labov Reference Labov, Lehmann and Malkiel1982:55; Winford Reference Winford1984). This could be because “grammatical change can affect an entire population, with adults and children acquiring the new form at the same time. This means that syntactic change would not be detected in a purely synchronic study” (Labov Reference Labov, Lehmann and Malkiel1982:68).
As the last paragraph shows, discussions of the scarcity of variation outside phonology have centered on variation in syntax. This is probably because—notwithstanding recent developments in Relevance Theory, which views pragmatics as a submodule of a mind-reading module dedicated to processing ostensive stimuli (Wilson Reference Wilson2005 pace Sperber & Wilson Reference Sperber and Wilson1986/1995:65–71)—pragmatics is often viewed as a matter of performance, external to knowledge of the language itself. The resulting absence of pragmatics from discussions of variation has, of course, done little to promote variationist pragmatic analysis. Nevertheless, the richness of potential loci of pragmatic variation outlined in the previous section makes clear that pragmatic variation is not infrequent. Moreover, the pervasiveness of “pragmatic failure,” that is, “misunderstandings which arise, not from any inability on the part of [the] H[earer] to understand the intended sense/reference of the speaker's words in the context in which they are uttered, but from an inability to recognize the force of the speaker's utterance when the speaker intended that this particular hearer should recognize it” (Thomas Reference Thomas1983:94) suggests that pragmatic variation is also quite salient—indeed, when pragmatic variation is present, the overall quality of interaction is directly affected (ibid.). The reason, then, for the relative neglect of the study of pragmatic variation from a variationist perspective must be sought elsewhere. I would like to propose as such a reason the lack of appropriate tools for analysis. Given the priority traditionally accorded to truth-conditional equivalence in defining linguistic variables, it is not surprising that variants of a single pragmatic variable have not been readily identifiable. Pragmatic variants are hardly ever truth-conditionally equivalent; nevertheless, they can be used to achieve equivalent perlocutionary effects in discourse (take, for instance, the possible linguistic realizations of a request to open the window). As such, the tools for analyzing pragmatic variation should be sought within pragmatic theory itself.
Conceptual vs. procedural meaning
The recent undermining of the truth-conditional/non-truth-conditional distinction in favor of a conceptual/procedural one (Wilson & Sperber Reference Wilson and Sperber1993; Carston Reference Carston2002:160–64) is worth exploring in this respect. Relevance Theory (Sperber & Wilson Reference Sperber and Wilson1986/1995) views truth-conditional meaning as a property not of linguistic expressions but of conceptual representations that are jointly produced by linguistic decoding and pragmatic inference (explicatures, whether basic-level or higher-order) or by inference alone (implicatures). What is important about linguistic expressions is how they contribute to this process. This can be done in two ways: linguistic expressions can directly contribute concepts, that is, constituents of conceptual representations, in which case they encode conceptual meaning; or they can contribute information about how to manipulate and combine these concepts into a conceptual representation. In the latter case, linguistic expressions encode procedural meaning (Blakemore Reference Blakemore1987, Reference Blakemore2002:78ff.; Wilson & Sperber Reference Wilson and Sperber1993; Carston Reference Carston2002:161–64).
The notion of procedural meaning was developed by Blakemore (Reference Blakemore1987) to account for the meaning of discourse connectives previously analyzed as conventional implicatures (Grice Reference Grice, Cole and Morgan1975/1989), which were both encoded (semantic) and non-truth-conditional (pragmatic). Wilson & Sperber summarize this proposal as follows:
discourse connectives such as ‘so’ and ‘after all’ should not be seen as encoding concepts. They do not contribute to the truth-conditions of utterances but constrain the inferential phase of comprehension by indicating the type of inference process the hearer is expected to go through. … such expressions contribute to relevance by guiding the hearer towards the intended contextual effects, hence reducing the overall effort required. (1993:11)
However, Wilson & Sperber (Reference Wilson and Sperber1993) go further than Blakemore in deconstructing the primacy of a truth-conditional account by suggesting that, in addition to discourse connectives, a large number of other elements—including indicative, imperative, and subjunctive mood indicators, rising or falling intonation, inverted word order, particles like please and let's, pronouns, and demonstratives—also encode procedural meaning. Moreover, some of these, such as pronouns and demonstratives, encode procedural meaning that guides the identification of appropriate referents entering into the truth-conditions of the basic-level explicature (= the proposition expressed par excellence in Relevance Theory), such that there can be no easy identification of procedural with non-truth-conditional meaning. Rather, the two distinctions are orthogonal: some expressions—for example, pronouns—encode procedures but affect truth-conditions, while others—for example, sentence adverbials—encode concepts but do not affect truth-conditions.
According to Relevance Theory, an utterance can achieve relevance in one of three ways: (i) it can enable the derivation of new contextual assumptions; (ii) it can strengthen existing assumptions; or (iii) it can lead to the elimination of existing assumptions (Sperber & Wilson Reference Sperber and Wilson1986/1995:121). Blakemore (Reference Blakemore1992:137–42) showed how discourse markers such as so, after all, and however encode procedures that can lead to any one of these types of cognitive effect. However,
not only is an analysis of procedural meaning based on the three cognitive effects of contextual implication, strengthening and elimination not fine-grained enough to capture the differences between expressions such as but and nevertheless and however, but it is also not broad enough to capture all the ways in which linguistic expressions and structures can encode information about the computations involved in the interpretation of the utterances that contain them. (Blakemore Reference Blakemore2002:98)
Consequently, Blakemore has elaborated the notion of procedural meaning in two directions. First, procedural meaning can constrain not only the direction of inference (the type of cognitive effect) but also the types of contexts in which the expression may be felicitously used. Thus, the difference between but and however lies in that, further to indicating the elimination of a previously accessible assumption, however additionally indicates that the speaker is not wholly dismissing the relevance of that assumption (2002:121); while the difference between these two connectives and nevertheless lies in that only nevertheless can be used in contexts where the elimination of this assumption amounts to accepting one way of putting things (the speaker's) over other, equally plausible ones, achieving a persuasive or “rhetorical” effect (2002:128). Accepting that “the information encoded by a linguistic expression or construction may activate either an inferential route or a particular kind of context or, indeed, both” (2002:128) allows us to capture both the similarity between these expressions, on account of which they constitute a “family of expressions” related to but, as well as the fact that they are not interchangeable in all contexts (2002:115).
Furthermore, procedural meaning may be considerably broader, and in effect not tied to any single one of the three types of cognitive effects outlined above. This is the case with the discourse connective well Footnote 15 that seems to encode a general instruction “to go ahead with the inferential processes involved in the derivation of cognitive effects” (2002:147). In such cases, well serves simply to re-affirm the presumption of optimal relevance communicated by every act of ostensive inferential communication, something that the speaker is likely to want to do when the relevance of her utterance is in question. Since, as Blakemore points out, “a speaker cannot be expected to guarantee the optimal relevance of another person's utterance or thoughts” (2002:127), this analysis accounts for the intuition that well “ties an utterance they preface [sic] to the current speaker” (Schourup Reference Schourup2001, quoted in Blakemore Reference Blakemore2002:145).
Finally, conceptual and procedural meanings are not mutually exclusive. Rather, as argued for phenomena such as grammaticalization (Nicolle Reference Nicolle1998), discourse connectives (Lee Reference Lee2002), mood indicators and illocutionary particles, and pronouns and sentence adverbials (Fraser Reference Fraser2006), linguistic expressions may encode both types of meaning simultaneously, and the balance between these two types of meaning may also shift over time. Pursuing this line of thought to its (natural?) conclusion, Sperber recently argued that:
All lexical items encode procedural content. Some also encode conceptual content. When a conceptual content is encoded, so is ipso facto the instruction to inferentially construct an ad hoc conceptual content taking the encoded conceptual content as a starting point of the inferential process. (http://www.phon.ucl.ac.uk/home/robyn/relevance/relevance_archives_new/0140.html; accessed June 5 2009)
Although the theoretical and empirical implications of this hypothesis remain to be worked out, this view is closest to the notion of sense modulation advocated in contextualist approaches (Récanati Reference Récanati2004; see Critiquesof the Labovianvariable and proposed solutions). In conclusion, procedural meaning is a powerful notion whose full explanatory potential has yet to be explored (Carston Reference Carston2002:216, n. 38). Its application to pragmatic variation in the next section is proposed as a step in this direction.
Procedural meaning as a basis for defining the pragmatic variable
The idea that the meaning of an expression may reside not in how it relates to states of affairs in the world but in instructing the hearer how to construct an interpretation of an utterance containing this expression is not new to Relevance Theory. Ducrot (Reference Ducrot1972:128–31) had advanced a similar idea within his theory of argumentation, which saw the argumentative function of language as prior to and guiding its informative function. Procedural meaning, aimed at reducing the search space for possible interpretations of the incomplete semantic representations that are the output of linguistic decoding, is by definition oriented not towards the world but towards the cognitive coordination between speaker and addressee. As such, this notion makes sense to incorporate into an account of language variation that prioritizes intersubjectivity, such as was briefly sketched above.
Remarks such as the following by Wilson & Sperber further support the compatibility of procedural meaning with the variationist enterprise:
As is well known, the functions performed in English by mood and word order are performed in many other languages by so-called discourse or illocutionary particles. Certain dialects of French, for example, have an interrogative particle ti, which appears to perform the same functions as word-order inversion does in other dialects. If word-order inversion is correctly analysed as encoding not a concept but a constraint on higher-order explicatures, then by the same arguments, illocutionary particles such as ti … should be analysed in similar terms. (1993:22, emphasis added)
This suggestion amounts to proposing that an instance of regional variationFootnote 16 in illocutionary force marking—in other words, an instance of pragmatic variation—may be analyzed in procedural terms, by considering the two regional variants to be encoding (at least in part) the same procedural meaning, despite one being a lexical item (the particle -ti) and the other a syntactic construction (word-order inversion). Then, on the model of Blakemore's (2002:115–28) analysis of the connectives but, however, and nevertheless, wherein all three constrain the inferential process in the same direction but the last two encode additional constraints on context that the former lacks, these constructions may be considered variants of each other, albeit not substitutable in all contexts.
The idea that variation beyond phonology should be sought across components of the linguistic system rather than within any single one of them has previously been suggested by Cheshire (Reference Cheshire2005). Cheshire analyzed the marking of discourse-new entities by British adolescents in a corpus of sociolinguistic interviews conducted in three English towns (Cheshire et al. Reference Cheshire, Kerswill and Williams1999). Her search for the linguistic forms that realize this marking in her corpus—the envelope of variation—led her to consider, in addition to existential clauses (involving left dislocation, possessive have (got)-constructions, and it-constructions), several other forms used in conjunction with the NPs introducing discourse-new entities, including tags (and stuff, and everything; cf. Dines Reference Dines1980), pragmatic particles (like, sort of, you know), high rising tones, indefinite this, uttering an NP and immediately expanding it, non-restrictive relative clauses, as well as hesitations, repetitions, filled and unfilled pauses, and false starts (Cheshire Reference Cheshire2005:485–89). However, rather than in the distribution of these forms, it was in the frequency of their use versus the absence of any kind of marking—that is, introducing discourse-new entities by bare NPs—that socially meaningful variation emerged in her data. Specifically, it turned out that “[i]n all three towns there was a highly significant gender and social class distribution, with working-class female adolescents using the highest proportion of bare NPs and middle-class male adolescents the lowest proportion” (2005:490).
Cheshire proposed a tentative explanation of this finding drawing on what is known about the macrosocial variables of gender and social class from previous research (2005:494–99). What is important for our current purposes is her conclusion that:
speakers of different genders and from different social classes may make different choices about when to select one (or more) of the available strategies to mark the introduction of a discourse-new entity. It would be instructive to perform more detailed qualitative analyses to attempt to determine why speakers mark their discourse-new entities on some occasions and not on others; but for the time being we can simply note that speakers do not, in fact, all act the same way: sociolinguistic variation exists at the pragmatic level as well as in other aspects of language. (2005:501)
This is very important because it empirically demonstrates the oft-repeated claim that meanings are not uniformly distributed across a population (Lavandera Reference Lavandera1978, Reference Lavandera, Dittmar and Schlieben-Lange1982; Dines Reference Dines1980; Garcia Reference Garcia1985; Eckert Reference Eckert2008; see Critiques of the Labovian variable and proposed solutions). Moreover, not only is Cheshire's analysis couched in broad variationist terms (quantitative measures, correlation with macrosocial variables) and framed as a case of “pragmatic variation,” but, I would argue, her results can be recast in terms of procedural meaning as outlined above. Specifically, the procedural meaning that binds together the forms listed above is an instruction to identify an NP as introducing a discourse-new entity. Socially meaningful variation emerged, in Cheshire's results, not so much in the different groups' use of the various forms encoding this procedural meaning, as in their degree of linguistic encoding of this instruction; that is, groups varied in their use of the variable itself.
The results of other recent studies can be similarly recast as cases of pragmatic variation involving procedural meaning. Rickford, Buchstaller, Wasow & Zwicky analyzed intensifier uses of all, some of which have been around since Old English, as well as quotative uses of all, which are much more recent, originating in California in the 1980s (2007:4). They define intensifier all as a variant of a variable that “includes every option speakers have at their disposition to reinforce or boost the property denoted by the head they modify” (2007:7), a function in which all competes with several other boosters such as really, very, totally, so, completely, such, way, and fuckin' (Rickford et al. Reference Rickford, Buchstaller, Wasow and Zwicky2007:8). This intensification function can be easily recast in procedural terms, whereby the instruction encoded by these variants is to interpret the property denoted by the head as one that the speaker wishes to foreground or reinforce. Moreover, their analysis revealed contextual constraints on the distribution of intensifier all. Using Dixon's (Reference Dixon1982) classification of adjectives into semantic types, they found that
All is most favored with adjectives denoting physical property and those denoting age, color and speed. … Therefore, while all has a more restricted distribution than the ubiquitous really …, its distribution is as broad as that of very, the prototypical booster, which is slightly higher in frequency. (Rickford et al. Reference Rickford, Buchstaller, Wasow and Zwicky2007:10–11)
These contextual restrictions can be captured as additional constraints on the possible contexts of occurrence of these variants, providing a more fine-grained analysis of their procedural meaning along the lines of Blakemore's (2002) analysis of the but-however-nevertheless hierarchy described earlier.
Coming to quotative uses of all, Rickford and colleagues subsume these under a variable that includes “all strategies used to introduce reported speech, sounds, gesture, and thought by self and other” (Buchstaller Reference Buchstaller2006, cited in Rickford et al. Reference Rickford, Buchstaller, Wasow and Zwicky2007:12). This is once again clearly a procedural definition that serves to bring under a common umbrella variant forms encoding the instruction to interpret the speech that follows as attributed to a speaker other than the speaker's current self. In this function, all competes with like, say, and go, as well as unframed quotes (paralleling Cheshire's bare NPs introducing discourse-new entities discussed earlier). The contextual constraints Rickford and colleagues discovered in the use of quotative all are:
the tense of the quotative, the nature of the quote (Speech/Thought), where the occurrence of outwardly realized speech favored all, as well as the Quotative Harmony factor group, where all was disfavored in contexts that did not contain any previous quotation. (2007:15)
This rather disparate group of syntactic, semantic and discourse factors can be captured as additional pragmatic constraints (i.e. instructions) encoded by quotative all with respect to its possible contexts of occurrence, and can help distinguish it further from associated variants expressing the same procedural meaning so far as the inferential route the hearer should follow is concerned. However, the authors' most interesting finding regarding quotative all is that:
in 1990 and 1994 all was the predominant variant among the younger high school students but was categorically constrained by age, being not used at all among the college-age speakers. Among the speakers who used all, it mainly introduced reported speech and occurred with present-tense verbs and in contexts of reported speech. However, by 2005, all had fallen back to below 10% among the quotative options, and the younger cohort used it even less than the college-age speakers. (Rickford et al. Reference Rickford, Buchstaller, Wasow and Zwicky2007:22)
In other words, the use of quotative all seems to be stratified by age as well as region,Footnote 17 that is, to constitute a case of socially meaningful pragmatic variation in the variants speakers of different cohorts choose to express the same procedural meaning.
As a final case of pragmatic variation that may be recast in procedural terms, I consider Williams & van Compernolle's (2009) investigation of the distribution of generalizing uses of the French pronouns on, tu, and vous with indefinite reference in a corpus of synchronous electronic discourse. As mentioned earlier, in Relevance Theory pronouns are analyzed as both procedural and truth-conditional, since “they guide the search for the intended referent, which is part of the proposition expressed” (Wilson & Sperber Reference Wilson and Sperber1993:21). However, Fraser (Reference Fraser2006:29) has argued that (at least) pronouns such as he also encode conceptual information, in this case the concepts of masculine and singular. If that is correct,Footnote 18 then the difference between definite and indefinite uses of on, tu, and vous may be analyzed as a difference in the ratio ofconceptual:procedural meaning encoded by each, with indefinite uses encoding the least conceptual meaning, if any. This ratio may be conditioned by linguistic and/or extralinguistic factors, making it an integral part of the pragmatic variation under study.
Williams & van Compernolle's results can be recast in these terms. Although the lack of demographic information about their subjects, given the anonymity of the medium, does not allow generalizations about the macrosocial variables investigated in previous studies that were based on sociolinguistic interviews (e.g. gender, class, and age; Williams & van Compernolle Reference Williams and van Compernolle2009:411), they nevertheless reach interesting conclusions that replicate those of earlier studies as well as add an interesting diachronic twist. Their results replicate earlier findings concerning the linguistic factors conditioning the choice between on and tu/vous to express indefinite reference, with generalizations, morals and truisms, nonspecific events, and negative utterances favoring on, and, conversely, implicatives, situational insertions, specific events, and affirmative utterances favoring tu/vous (2009:423–25). In this case, the former set of factors can be said to constitute contextual constraints tipping the ratio of conceptual:procedural meaning towards procedural meaning for on, while the latter set of factors constitute contextual constraints tipping the ratio of conceptual:procedural meaning towards procedural meaning this time for tu/vous.
Compared with previous studies, Williams & van Compernolle's results showed an increased preference for indefinite on vs. indefinite tu/vous (2009:423). In this regard, it is useful to recall that, in its definite uses (for 1st person plural reference), on is the colloquial variant alternating with nous ‘we’ as the standard one, while in its indefinite uses, on is the standard variant alternating not only with tu/vous but also with ils ‘they’, all of which have been proscribed at one time or another (2009:410). Definite on, in other words, is a Labovian stereotype, witness the old adage on c'est un con ‘on is a jerk.’ This state of affairs may well be affecting the stratification of indefinite on by age reported in previous, interview-based studies, which showed speakers over thirty favoring indefinite tu/vous (the more colloquial variants), while younger ones, who are of school age and presumably more susceptible to prescriptive pressure in an interview situation, favored indefinite on (the more standard one). In other words, the meaning of on seems to be different for these two age cohorts, with on expressing a stronger conceptual (1st person reference) component for older speakers, resulting in their preference for indefinite tu/vous instead, and a stronger procedural (indefinite reference) component for younger ones.
While the heightened preference for indefinite on they found may thus be explained as the tail-end of a change in progress,Footnote 19 Williams & van Compernolle additionally suggest two explanations specific to the medium of synchronous chat. First, they point to the lack of face-to-face contact as a possible reason compelling speakers to disfavor indefinite tu/vous in favor of indefinite on in their wish to avoid potential confusion between the former and definite uses of tu/vous (2009:425). Second, they propose an explanation that appeals to the variants' functional load in this medium. As informal tu is becoming generalized for definite (2nd person) reference in synchronous chat, winning over its formal competitor vous, this “results in its decreased capacity to serve also as an indefinite pronoun” (Williams & van Compernolle Reference Williams and van Compernolle2009:423), causing the increased percentage of indefinite on in the synchronous-chat study when compared with the earlier interview-based ones. Again, this explanation can be recast in procedural terms, whereby the ratio of conceptual:procedural meaning encoded by each variant is shifting according to generational cohort and medium of communication. Notice, however, that to reach this explanation, we had to consider both those contexts in which the variants alternate (with respect to expressing indefinite reference) as well as those in which they do not. All of these contexts can be captured under an analysis that acknowledges the shifting ratio of conceptual:procedural meaning encoded by each of these forms and proceeds to specify the differences between them as additional constraints on the particular inferential route that they activate, or the particular context, “or indeed both” (Blakemore Reference Blakemore2002:128).
In this section, I revisited three recent studies of pragmatic variation and showed how their results can be recast in procedural terms. Both types of pragmatic variation outlined earlier (see Variational pragmatics and the study of pragmatic variation) were illustrated: a single procedural meaning encoded by different forms (Cheshire Reference Cheshire2005), and a single form encoding different procedural meanings (Rickford et al. Reference Rickford, Buchstaller, Wasow and Zwicky2007). Moreover, I showed how pragmatic variation can involve, not the particular procedural meaning encoded by a specific form, but rather the ratio of conceptual:procedural meaning encoded by a linguistic form in different contexts (Williams & van Compernolle Reference Williams and van Compernolle2009). The analyses in this section thus go beyond Schneider & Barron's (2008) “formal level of analysis,” indicating ways of additionally analyzing variation at their actional, interactional, topic, and organizational levels using the proposed procedural conceptualization of the pragmatic variable.
CONCLUSION
The linguistic variable was proposed by Labov in the early 1960s as a way of bringing together and studying under a common umbrella items of linguistic form that alternate in a socially meaningful way. Given its inception to study phonological variation, it should not come as a surprise that its original definition incorporated several features that seemed to be part and parcel of phonological variants—increased frequency, immunity to total suppression, and not least, truth-conditional equivalence—but proved more or less problematic when it came to extending the variable to non-phonological variants. Therefore, the question that this article set out to address is: Can the linguistic variable be extended to account for variation in pragmatics? If so, how?
I have proposed an extension of the linguistic variable that has proved so productive in the analysis of phonological variation with the aim of accounting for variation in pragmatics, understood as both variation in the range of forms expressing a particular meaning, and variation in the meanings expressed by a single form. This extension builds on the notion of procedural meaning, developed within Relevance Theory to capture the constraints that linguistic forms place on the inferential process, even though the products of this process (the actual truth-evaluable propositions derived, be they explicatures or implicatures) may be different each time. Under a procedural definition of the pragmatic variable, what remains stable is the process of inference (the instructions encoded in a linguistic form), and it is on the basis of this that forms may alternate and be considered variants of each other. Procedural meaning thus provides an opportune way of operationalizing in the case of pragmatic variation the notion of functional equivalence, which is widely accepted as an improvement over the problematic assumption of truth-conditional equivalence of linguistic variants. Since the notion of truth-conditional equivalence followed from an emphasis on the representational function of language, it is not a coincidence that, as the emphasis is shifted towards the argumentative function of language, procedural meaning, aimed at the cognitive coordination between speaker and addressee, should provide the basis for the definition of pragmatic variables.
Of course, capturing the quantitative dimension of pragmatic variation does not necessarily mean adopting the notion of the linguistic variable and the broader theoretical framework within which it is couched. It is quite possible that there are other ways of unravelling quantitative aspects of language use that have yet to be discovered and are more appropriate to the analysis of pragmatic variation. To some, the definition of the pragmatic variable in procedural terms proposed in this article may already appear to be such an alternative, rather far removed from the original Labovian concept. Although this is true to some extent, and the proposed reconceptualization of language (and hence, of language variation) as serving first and foremost an argumentative function clearly not envisaged by Labov himself, I would argue that this way of interpreting the pragmatic variable is nevertheless in consonance with Labov's original functionalist conception of linguistic variables as socially motivated, hence transient, clusterings of forms (Labov Reference Labov1966:16). As such, what I am proposing is an extension of the variationist program rather than its abandonment. It remains to be seen to what extent this proposal will prove fruitful for the analysis of pragmatic variation and how it will be revised in the light of future studies. At the same time, I believe that explicit, theoretically informed dialogue between the fields of sociolinguistics and pragmatics, such as I have tried to initiate with this article, is long overdue and one from which both fields stand to profit. I can only hope others will share this view.
