18 Sociolinguistics and Formal Linguistics

Gregory R. Guy

18.1 INTRODUCTION theories of this faculty are inadequate unless they account for this. Sociolinguistics is often seen as having little rele- One contribution of sociolinguistics to vance to formal linguistic theory. Indeed, a formal formal linguistics is therefore observational and linguist once told the present author that sociolin- descriptive – it turns up facts that theory must guistics ‘is not linguistics’. While such a comment account for. Formal models sometimes flee from reveals a narrow and sectarian view of linguistics, the seemingly disorganized multitude of such it also shows that the relationship between social facts and declare themselves not responsible for aspects of language and formal models of lan- the mess. This is an appropriate initial scientific guage structure is not self-evident – at least, given response to complexity – focusing on a subset of prevailing theoretical frameworks in the field. So facts, idealizing them and ignoring interactions why does this volume have a chapter addressing with other facts. But when disciplines mature to this relationship? the point of achieving some mastery over their The most obvious connection lies in the funda- subject matter, the gains from idealization are mental scientific concern with theoretical ade- outweighed by the concomitant limits on the the- quacy. Chomsky (1964: 29) identifies the lowest ory’s capacity to achieve a more accurate account level of adequacy as ‘observational’ – providing a of reality. At this point a serious science is obliged faithful representation of the data. Formal linguis- to engage with what was previously ignored. tic theories that ignore social diversity – a demon- Linguistics has now reached this phase, and strable characteristic of all languages – do not formal linguistics has begun to recognize the need meet this most elementary criterion, let alone the for a sounder empirical basis, and to seek broader highly valued descriptive or explanatory adequacy empirical testing of theory, which can only be they aspire to. Formalists have glossed over this achieved in the messy world of language use. glaring lacuna by addressing their models to ever- Sociolinguistic studies thus provide a proving narrower social universes, ultimately the ‘idiolect’ ground for theory that is unavailable in the intro- – the grammar of a hypothetical monostylistic spective, intuitive paradigm of theoretical invariant individual. But observation denies even research. this refuge, demonstrating that such an object Sociolinguistic research also points towards does not exist in the world. And even if it did, a what is needed in linguistic theory to achieve formal model that accounted for only this would higher levels of adequacy. Adequate formal models be limited to production, leaving the other two must eventually account for variability – an indi- pillars of the language tripod – perception and vidual’s capacity for stylistic variation, for accom- acquisition – unaddressed, since human beings modation, for performing acts of identity by encounter sociolinguistic diversity in the speech linguistic means – But what are the limits and the of others, and must be able to learn language and landscape of that variability? An adequate formal understand others even when they speak differ- theory must incorporate social information – ently. Hence the obvious conclusion is that the social interpretations of variables and ways of human language faculty acquires, generates, rec- speaking; But what kinds of social information are ognizes and interprets variability, and that formal linguistically relevant? Sociolinguistic work is

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1 33/22/2010/22/2010 6:11:266:11:26 PMPM 2 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

developing the outlines of the necessary elements theory – the Constant Rate Hypothesis – that of an adequate theory of language – one that states that changes move at the same rate in all accounts not only for narrow Chomskyan compe- contexts, and are therefore not driven by contex- tence but also, following Hymes (1972; see tual effects. Guy (1991, 1992) applied the Lexical Johnstone and Marcellino, this volume), for speak- Phonology model of Kiparsky (1982a) to variable ers’ sociolinguistic or communicative compe- coronal stop deletion (CSD) in English, deriving tence, their capacity to speak and understand in a predictions of an exponential relationship among diverse and varying linguistic world. Encouragingly, retention and deletion rates in three different mor- this work is meeting complementary develop- phological categories, which has been confirmed ments from the formal side, in theoretical frame- in studies of several English dialects (cf. Santa works such as Optimality Theory (OT) and Ana, 1992; Bayley, 1994). usage-based phonology. Some of the formal models in sociolinguistic The other side of this relationship is the impor- work have pioneered in areas where other formal tance of formal linguistics to sociolinguistics. theories have feared to tread. Sociolinguists have Sociolinguists have, unfortunately, sometimes been at the forefront of work on interlingual adopted a reciprocally sceptical view of the rele- issues, notably the formal models of code-switch- vance of formal theory to their subdiscipline. ing proposed by scholars such as Poplack (1980), Ironically, their reasons for doing so are similar to Myers-Scotton (1993) and Nishimura (1997). formalists’ reasons for neglecting sociolinguistic Formalists have done little work on linguistic results: the models of formalists are diverse, structure above the level of the sentence, but messy and constantly in flux. The impression that sociolinguists have proposed formal models of outsiders sometimes have of formal linguistics is discourse level phenomena such as narrative reminiscent of T. S. Eliot’s famous poem, Prufrock: structure (Labov and Waletzky, 1967), conversa- ‘In the field, the theories come and go, talking of tional turn-taking (Sacks et al., 1974) and thera- Michelangelo; there is time for a hundred visions peutic discourse (Labov and Fanshel, 1977). Thus and revisions before the taking of the toast and the relationship of sociolinguistics to formal tea’. If theories are ephemeral, focus on irrelevan- theory comes full circle, with sociolinguists not cies, and change a hundred times before tea, an only following up the analyses suggested by empirical researcher might be wise to avoid formal models but also testing and revising them, making them a central concern of a study that and proposing new ones. might take years to complete. Of course, as this volume makes clear, sociolin- Despite sometimes holding such attitudes, soci- guistics has become a very broad discipline that olinguists cannot avoid theory when they analyse encompasses a considerable variety of interests data. Whenever they formulate a hypothesis, or and approaches, and not all of these will have the even identify a variable, they are relying on some same level of engagement with formal linguistics. model of events, some ‘theory’ (perhaps implicit) Studies of language policy and planning, or lan- of the structure of language. When Labov famously guage endangerment, are marginal to the con- studied coda /r/ in New York City (1966), he made struction of formal models of language, whereas a formal claim to the effect that /r/-less and /r/-ful sociolinguistic research that examines detailed pronunciations of words like source, car, beer contextual constraints on phonological or syntac- alternated in the speech of New Yorkers, and that tic variation is intimately involved with analysing the alternants counted as lexically equivalent even linguistic structure. But subdividing sociolinguis- though they were phonologically and socially dif- tic interests into those that engage with formal ferentiated. This analysis rested on theoretical theory and those that do not is not necessarily constructs such as phoneme, word, equivalence, simple or clear. Consider work on language con- difference and phonological alternation. In gen- tact. The Franco-Norman conquest of England in eral, theoretical models suggest to the sociolin- 1066 is a sociohistorical fact with no direct con- guist what things might be interesting to nection to linguistic theory. But the massive bor- investigate. A great deal of sociolinguistic research rowing of French and Latinate words into English has been devoted to investigating and testing that ensued has complicated English phonology in hypotheses derived from theoretical models. ways of considerable interest to formal linguists. And much sociolinguistic work seriously Much theoretical work in phonology addresses engages with the construction of formal models. problems of this sort, such as how to associate Bailey (1973) proposed a contextually-driven certain phonological patterns with specific subsets theory of variation and change, in which change is of the lexicon. Similarly, code-switching was first hypothesized to begin earliest and proceed fastest studied by sociolinguists interested in the social in maximally favourable contexts. Kroch (1989b) phenomenon of alternating languages in bilingual tested this model against data on syntactic change communities. But it has attracted much formal and disproved it, proposing instead an alternative theoretical attention, due to the insights it offers

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 2 33/22/2010/22/2010 6:11:276:11:27 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 3

into questions of syntactic structure and linguistic justified by a formal model in which words were cognition. composed of phoneme-like units on which sound Therefore we cannot, a priori, partition socio- change operated (Paul, 1978 [1886]). When such a linguistics into theoretically relevant and theoreti- unit changed, all words containing it changed cally uninteresting components. All its subdivisions simultaneously. Apparent ‘exceptions’ to these cor- have the two connections to linguistic theory we respondences were postulated to result from proc- have identified: all have the descriptive connec- esses like borrowing and analogical change that tion, revealing facts that theories may eventually operated after the categorical change occurred. seek to explain, and all rely on hypotheses about Sociolinguistic and dialectological observations of language structure that are ultimately derived variability were relegated to the periphery of this from theoretical models. Since we assume, in research paradigm, by definition; they were noise principle, that language cannot be adequately obscuring the systematic big picture. (See also understood in isolation from its social embedding, discussion in Kerswill, this volume.) we assume a general relevance of sociolinguistics The Neogrammarian worldview was motivated to formal theory. in part by the discovery of phonological condi- In what follows, this chapter considers the fol- tioning. For example, Verner (1978 [1877]) lowing topics: how theory informs sociolinguis- famously discovered a prosodic constraint on tics, how sociolinguistic studies address formal Grimm’s Law: voiceless stops following an issues, and current developments in formal theory unstressed vowel in Proto-Indo-European are that constitute progress towards an adequate voiced in Germanic, instead of becoming voice- model of language – one that has some possibility less fricatives. Such discoveries provided a para- of accounting for the sociolinguistic facts. But digm for formal analysis in linguistics that persists before addressing these topics, we begin with a in some measure to the present day: search for brief survey of the history of linguistic theory invariant generalizations, ignoring variability insofar as it has spoken to, or ignored, sociolin- present in usage; where persistent exceptions to guistic questions. those generalizations are encountered, search for other invariant sub-generalizations delimited by formally definable contexts. This has been a productive heuristic for linguists, but it has 18.2 PAROLE, PERFORMANCE, also provided formal linguistics with an oubliette PERIPHERY: THE for disposing of inconvenient facts: one postulates MARGINALIZATION OF that any exceptions will be accounted for in the future by some more refined statement of the SOCIOLINGUISTICS IN context. At worst, this is unfalsifiable: all formal FORMAL THEORY processes are exceptionless, and any observable exceptions will be explained by future Verners. Ever since the Neogrammarian movement revolu- For linguists who take a Popperian view of tionized linguistic theory in the late nineteenth their science, such a hypothesis need not be taken century, the mainstream of formal thinking about seriously. language has pushed sociolinguistic questions The invariant view of linguistic process was either to the margins of the discipline, or beyond adapted to synchronic theory by Saussure, the the pale. The Neogrammarians provided a key founder of modern formal linguistics. Saussure idealizing assumption that justified this marginali- (1916) elevated the separation of social and zation: the concept of invariance in linguistic proc- formal elements of language to a theoretical prin- esses. The dominant focus of linguistics in the ciple via his dichotomy between langue and nineteenth century was historical and comparative, parole. For Saussure, langue is the abstract, sys- leading to the discovery of numerous phonological tematic aspect of language, while the social diver- correspondences between related languages, such sity of language use is assigned to parole. These as the ‘Grimm’s Law’ relationship between have unequal status in linguistic science: Saussure Germanic and Proto-Indo-European (roughly privileges langue as the object of study of serious speaking, Germanic underwent a chain shift of the linguistics. Sociolinguistic concerns are consigned form: /bh, dh, gh/ > /b, d, g/ > /p, t, k/ > /f, θ, h/). to the periphery of Saussurean thought, as the Many exceptions to such correspondences were study of parole. well known, and an emergent dialectological tradi- Saussure’s dichotomy is recast by Chomsky tion was documenting extensive variability in lexi- as competence and performance, and later, cal development. But despite the empirical evidence I-language (internalized) and E-language (exter- of variability, the Neogrammarians argued that nalized). Competence/I-language is the mental correspondences arose from EXCEPTIONLESS sound linguistic system governing production – the change. The hypothesis of ‘exceptionlessness’ was mental grammar. It is assumed to follow invariant

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 3 33/22/2010/22/2010 6:11:276:11:27 PMPM 4 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

principles in the Neogrammarian ‘exceptionless’ full of variety, it is reasonable to postulate that tradition. Performance is the product of the system linguistic systems – mental grammars – generate, – actual speech, or any other externalized linguis- perceive and interpret variability. Since no tic output (hence E-language). Mediating between Chomskyan invariant ‘ideal speaker-listener’ the mental grammar and its products are psycho- exists, why postulate such a person as the object logical and physiological factors that do not share that our theories seek to explain? Indeed, WLH go the property of invariance; hence, they introduce further; they argue that if such invariant speakers variability into performance. Chomsky cites fac- did exist, they would be social monsters, unable to tors such as slips of the tongue, memory lapses adapt to changing circumstances, or accommodate and performance errors as common flaws in eve- to different interlocutors, or to use language to ryday language use, and argues that such flaws are communicate all the social messages performed not the product of the mental grammar and do not by variant means. provide useful evidence about its nature (1965). The second WLH principle is ‘orderly hetero- The competence/performance distinction has geneity’: linguistic variability is not random and logical elegance, opposing a mental machine to its unstructured, the product of errors. Rather, it is products. But in the give-and-take of linguistic deeply structured by social and linguistic contexts. argumentation, it becomes the neutron bomb of Essentially, this counterposes a probabilistic inter- generative theory, removing humans from the bat- pretation of orderliness to the categorical interpre- tlefield, leaving only naked linguistic structures tation prevailing in formal theory. If system and behind. Sociolinguistics is collateral damage in order are equated to exceptionlessness, all valid this battle; it provides no data relevant to the con- generalizations must be categorically true; in struction of formal theory, because it looks only at quantitative terms, they must apply to 100% of performance. Chomsky specifically argues that cases. But a probabilistic model allows valid gen- ‘observed use of language … surely cannot con- eralizations about values below 100%. Social stitute the actual subject matter of linguistics, if groups are rarely distinguished by the categorical this is to be a serious discipline’ (1965: 4). presence or absence of a particular form, but they Performance is thus an even deeper oubliette for are often characterized by particular rates of use; the disposal of empirical counter-examples: data thus, in New York City, middle-class speakers use from usage bears unknown deviances from the more coda [r] than working-class speakers. And mental grammar, and therefore cannot contradict orderly heterogeneity is also evident in linguistic theories about that grammar. structure: many sociolinguistic variables show Chomskyan thought and its theoretical succes- strong probabilistic – but not categorical – condi- sors thus continue the century-old tradition of tioning by linguistic contexts. Thus, English final consigning sociolinguistic concerns to the periphery. CSD is much more likely preconsonantally (eas’ Linguists interested in the topics addressed in this side) than in prevocalic position (eas’ end). Nobody volume therefore find themselves with just two does this categorically, with 100% deletion before options for relating to linguistic theory. They can consonants and 0% before vowels, but every accept that their work addresses ‘mere’ perform- speaker’s usage shows probabilistic favouring of ance, and is consequently irrelevant to formal preconsonantal contexts. This process always has theory, or they can problematize the assumptions ‘exceptions’, but it is still systematic and orderly. that lead to this conclusion, most importantly the The challenge to formal linguistics that WLH assumption of invariance and the postulated oppo- present, therefore, is to design theoretical models sition between competence and performance. One that accommodate variability and probabilistic encounters, in sociolinguistics, work that follows orderliness. Formal models that continue to adhere both of these paths. One trend eschews attention to the assumption of invariance and the marginali- to theoretical issues – a position consistent with zation of parole/performance cannot account for the Chomskyan view that evidence from language the results emerging from sociolinguistic research, use is not, in principle, relevant to formal theory. and run the risk of ultimate irrelevance. But many Another, perhaps dominant, trend in sociolin- models are emerging that take up this challenge; guistic research has been to reject the sectarian some of these are discussed in subsequent sections. definition that excludes sociolinguistic concerns and denies their relevance to formal models of competence, and instead proposes alternatives that embrace variability and social diversity. This 18.3 FORMAL THEORY INFORMS approach found an influential formulation in the SOCIOLINGUISTICS work of Weinreich, Labov and Herzog (1968, henceforth WLH), which enunciates two princi- The fundamental finding of sociolinguistic ples that contradict the assumption of invariance. research is that variation and diversity permeate First is ‘inherent variability’: since language use is language. Individual speakers vary their usage

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 4 33/22/2010/22/2010 6:11:276:11:27 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 5

constantly and adaptively, to construct social iden- traits of a linguistic variable: it is conditioned by tities, demonstrate attitudes and manage social linguistic contexts, stylistically constrained relationships. Collectively, speakers are differenti- (speakers delete less in more careful styles) and ated by geography and time, and by social dimen- socially stratified (higher-status speakers delete sions of class, sex, ethnicity, age and linguistic less). But the linguistic constraints affecting it experience. Therefore, the fundamental questions have a curious interaction with the social con- in sociolinguistic research are: What is it about straints: some constraints are constant for all language that varies, and why? speakers of English, while others vary from dia- Because of the nature of the field, some answers lect to dialect. Why is this so? to the ‘Why?’ questions will be linguistic, and Let us consider two constraints affecting this some social and historical. Formal linguistics has process. First, as noted above, CSD is markedly great relevance to the former, providing potential affected by following phonological context, with explanations of constraints on variation. It will not significantly more deletion in preconsonantal than necessarily contribute to understanding the latter: prevocalic position. Also, the process is sensitive nothing about the structure of language accounts to morphology: stops representing the -ed suffix in for the social dominance of Spanish over Mayan regular verbs are deleted less often than those that in Guatemala, or for the spread of Northern Cities are part of the root morpheme. Relevant values dialect features in American English from larger from several studies are found in Table 18.1. to smaller urban areas (Callary, 1975). But it is The effect of following vowel (V) and conso- often difficult to distinguish one line of explana- nant (C) is, as the table indicates, systematic tion from another, and they may interact (if, for across all dialects. But, strikingly, the effect of example, the linguistic markedness of forms limits following pause (P) varies considerably from borrowability). Therefore, sociolinguistics must group to group. For some speech communities, allow the possibility of formal linguistic accounts pause is the most conservative environment, disfa- of any patterns of variation it encounters. vouring deletion even more than vowels, while for The most important contribution of formal others it is associated with high deletion rates. The theory may lie in answering the ‘What?’ ques- first discovery of this dialect-specific effect of tions. To make progress in appreciating the work- pause was made by Guy (1980), who found that ings of sociolinguistic diversity, we require New Yorkers favoured deletion before pause, concepts and hypotheses about the structure of while Philadelphians disfavoured it. The effect language – a ‘theory’ that suggests what things are was systematic within each speech community, as worth studying and where to look for them. Table 18.2 illustrates. Everybody in both commu- Clearly, our investigations are best served by nities (except two Philadelphians with small ns) bringing the best theories available to bear on the has more deletion before C than V, but 18 of 19 issues. If we want to know how speakers differ, or Philadelphians show pause as the least favourable what linguistic items they respond to in formulat- environment for deletion (P deletion rates ing social evaluations, we would do well to turn to below V), while no New Yorkers have this ranking formal theories to delineate the elements and of vowel and pause. processes that may be relevant. A similar issue arises with respect to morphol- As an example of the theoretical contributions ogy. Most speakers in most dialects systematically to sociolinguistic research, consider the variable show higher deletion rates in monomorphemic process of final coronal stop deletion (CSD) in (M) words like mist, pact, bold, vs reduced rates English, a coda-simplification process with paral- in regular past tense verbs (R) like missed, packed, lels in many other languages. It shows classic bowled. But speakers vary considerably in their

Table 18.1 Effects of morphological class and following segment on English coronal stop deletion Morphological Following segment MIRCVP Bayley 0.53 < 0.62 0.36 0.73 0.27 <0.46 Santa Ana 0.55 > 0.43 0.28 0.62 0.33 >0.32 Lim and Guy 0.60 < 0.65 0.26 0.61 0.37 <0.48 Guy 0.64 > 0.55 0.32 0.66 0.19 <0.37 M, monomorphemic words; I, irregular past tense verbs; R, regular past tense verbs; V, vowel; C, consonant; P, pause. Source: (from Bayley (1999), Guy (1991) Lim and Guy (2003) and Santa Ana (1992))

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 5 33/22/2010/22/2010 6:11:276:11:27 PMPM 6 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

Table 18.2 Comparing constraint effects in two speech communities: following context constraint orderings for coronal stop deletion in New York and Philadelphia Following context Number of speakers showing the order: C>V V>C V>Pause Pause>V Philadelphians 16 2 18 1 New Yorkers 4 0 0 4 C, consonant; V, vowel.

treatment of the irregular past tense verbs (I) that phonology (Clements and Keyser, 1983), and in take both a root vowel change and a coronal stop postulated OT constraints such as NoCoda, affix in the past, such as lost, kept, told. Guy and *CxCoda, Onset, etc. Hence, these theories sug- Boyd (1990) found that the treatment of this mor- gest a prediction, to the effect that … VCCCV… phological class varied with age: older speakers sequences, in phrases like east side, are univer- treat irregular verbs conservatively, with lower sally more marked than … VCCV … sequences, deletion rates, while younger speakers are more like east end. This being the case, a process that likely to delete them (Figure 18.1). simplified the coda sequence should always be Such findings provoke an obvious question: more strongly motivated in the former than the Why are some constraints constant across speak- latter. If these are linguistic universals, we should ers and dialects, while others are inconstant, either not expect to find any dialect or speaker that con- dialect-specific or age-graded? Since sociolin- tradicts them, which, essentially, is what the data guistics is simultaneously concerned with linguis- show – everybody deletes more before consonants tic structure and social context, we might entertain than before vowels. both social and linguistic explanations of these But what do formal models say about following facts. On the social side, a possible hypothesis is pause? Eligible words for English CSD have the that speakers arbitrarily select linguistic items for form …VCC, so when no other subsequent social-indexical purposes. This is akin to the arbi- segments are considered, they will violate NoCoda trariness of the linguistic sign: just as a book does and *CxCoda constraints just like eligible words not intrinsically demand the label book, since it is occurring in other contexts, but are they more or labelled hon in Japanese and kitab in Arabic, so less marked than prevocalic or preconsonantal there is no intrinsic reason that pronouncing coda contexts? Formal theories are largely silent on /r/ is socially favoured in New York. Therefore, we this point; an answer would depend on a formal might suppose that the prepausal deletion rate for analysis of non-speech (What are the distinctive coronal stops is arbitrarily selected as a dialect features of silence?), or on an additional set marker, and the deletion rate in irregular verbs has of hypotheses about syllabification, stop release, been arbitrarily associated with particular age etc., which would not be universal in nature. cohorts. Therefore, it is a fair conclusion from phonologi- Pending a more comprehensive analysis of cal theory that the effects of prevocalic and what variables and constraints have what social preconsonantal contexts are determined by lin- interpretations and associations, we cannot rule guistic universals, but the effect of following out this account on purely social grounds. But the pause is not universally defined, and hence is fact that numerous studies of CSD show the same available for dialect-specific, dialectally arbitrary effects for many other constraints, leaving just treatments. these two contexts to vary greatly across speakers, Formal theory suggests a similarly ambiguous should give us pause. Perhaps some linguistic status for the treatment of CSD in irregular verbs. property makes these constraints more amenable The difference between monomorphemic and to social differentiation. regular past tense forms is modelled in several This is where formal theory comes to our aid. ways in different theories: by functional con- Many strands of phonological theory agree that straints against loss of information; by a historical there is a universal hierarchy of syllable types, constraint against loss of morphological distinc- with CV being the most basic, found in all lan- tions (Kiparsky, 1982b, the ‘distinctness condi- guages, while other structures, including complex tion’); and by the different derivational histories codas, complex onsets and zero onsets, are of the two classes (underived vs derived, cf. Guy, more marked and typologically less common. 1991). But all models assign a distinct status to These principles are explicitly incorporated in CV the morphological marker in missed, bowled

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 6 33/22/2010/22/2010 6:11:276:11:27 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 7

that is absent in mist, bold, with a consistent same deletion rates as monomorphemes, which prediction that deletion of the past tense forms also lack internal morphology. Finally, in adult is disfavoured. life, many speakers proceed to a complex analysis But what about irregular past forms? There is in which the final coronal stop acquires a separate no obvious universal explanation for how they morphemic analysis, no doubt by analogy with the should be treated by a deletion process. The final regular affix in missed, bowled (thus keep~kep#t); stops in left, told mark tense redundantly, so they hence they begin to partake of the morphological have no greater functional load than those in resistance to deletion that regular past forms monomorphemes. Their derivational histories exhibit. depend on specific hypotheses about what form In both of these examples we find that theoreti- classes exist and how they are stored and gener- cal analysis supplies a principled basis for distin- ated, on which there is little theoretical consensus. guishing the constant constraints from those that Therefore, we can again entertain a reasonable vary dialectally or generationally; some elements hypothesis that while the difference between of linguistic structure are universal, and cannot monomorphemes and regular past tense forms is vary between speakers, dialects or languages, subject to universal conditions, the status of the while others are not universally specified, and irregular forms is precisely irregular, subject to hence are available for social differentiation. different analyses and different treatments by the This conclusion is supported by other studies of deletion process. sociolinguistic variation, beyond phonology and This is what Guy and Boyd (1990) argue. They beyond English. A case in point is Cameron’s see the age-grading in Figure 18.1 as a product of (1993) study of subject pronoun expression in successive reanalyses of these words by English Spanish. Spanish is a PRO-drop language – speakers as they develop linguistically. The young- subject pronouns are optional. In the meaning est (pre-teen) speakers classify them as strong ‘I want’, a Spanish speaker can say either Yo verbs, lacking a suffix (thus keep~kep, analogous quiero or Quiero with the pronoun Yo ‘I’ expressed with feed~fed); hence, final stops are almost com- or omitted. Cameron reports a major dialect dif- pletely absent. By adolescence, most speakers ference between Puerto Rico and Spain: Puerto construct a special morphological class for such Ricans use many more overt pronouns than their words, combining traits of the strong and weak Spanish counterparts. He also identifies linguistic classes, but without any internal morphological constraints on this process, including switch refer- analysis (thus keep~kept). Hence, they show the ence: speakers are more likely to use an overt

0.90

0.80

0.70

0.60

0.50 Prob. of –t,d absence

0.40

0.30 0 10203040506070 Age

Figure 18.1 Probability of –t,d absence in irregular past-tense verbs, by age (from Guy and Boyd, 1990: 8.)

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 7 33/22/2010/22/2010 6:11:276:11:27 PMPM 8 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

pronoun to mark a change in subject between the categorical – unsurprising, given the categorical previous clause and the current clause. Most lin- bias of linguistic theory. A case in point is the guistic constraints are common across the two Obligatory Contour Principle (OCP), a generali- dialects, but one is treated differently: specificity zation in phonology to the effect that languages of reference (the difference between references to prefer alternations (or ‘contours’) over repetition specific individuals and generic references mean- of identical elements. Thus the preferred structure ing ‘somebody, one, people in general’). For the for syllables and words involves alternating con- second-person singular (2sg) pronoun tú, this con- sonants and vowels, not strings of one or the straint is significant in both countries, but with other: CVCV over CCC or VVV. Similarly, tone opposite effect. Puerto Ricans use overt tú more languages prefer alternating tones, not strings of often when making a generic reference, while identical tones: HLHL, not HHH or LLL. Indeed, Madrileños favour overt tú for specific reference the OCP was first proposed by Leben (1973) to (Table 18.3). account for categorical tonal alternations in What does linguistic theory say about these African tone languages. constraints? The switch-reference effect has a A common finding of sociolinguistic research, clear functional motivation: the low information and of other quantitative studies of language, alternative (with no overt pronoun) serves as a gap however, is that essentially the same principles are onto which earlier references are projected. But observed in linguistic variation, but in a probabil- the higher information form, with an overt pro- istic, rather than categorical form. This discovery noun, is used to signal a new subject. English uses has been termed the ‘stochastic generalization’ contrastive stress in much the same way. In a sen- (Bresnan et al., 2001; Clark, 2005): generaliza- tence like ‘Billy talked to Paul about it, and HE tions that may be categorically true in one lan- said … ’ the stressed ‘HE’ refers to Paul, as a new guage exist as quantitative constraints on variation subject, whereas an unstressed alternative ‘… and in another language or social variety. he said … ’, would ordinarily indicate a continua- This is the result Guy and Boberg (1997) find tion of the same subject, i.e., Billy. In each case, for the OCP in their research on English CSD. The greater emphasis on the subject (via an overt pro- segment preceding a deletable final stop has a noun or contrastive stress) signals a switched ref- proven constraining effect on the likelihood of erence. This is a possible universal. But the deletion; roughly speaking, obstruents promote specificity constraint has no such universal asso- deletion more than sonorants. Guy and Boberg ciation with overt or zero pronouns. English, accounted for this effect in terms of the similarity which only allows overt pronouns, uses the 2sg between the preceding context and the deletion pronoun for both meanings. A sentence like ‘You target (the final coronal stop): greater similarity know what it means’ can mean either specifically (measured in terms of shared distinctive features) ‘you (the hearer) know … ’, or generically ‘one favours deletion. The values appear in Table 18.4. knows … ’. This suggests no intrinsic association The results show that segments that share two of overt pronouns with either meaning; rather, of the defining features with [t, d] all promote such an association is arbitrary, and two Spanish deletion more than those that share only one fea- dialects have made different selections for indicat- ture, while preceding vowels, which are com- ing genericity and specificity. pletely dissimilar, almost never trigger deletion. Note further that adjacent identical stops are cat- egorically prohibited in English (*tt, *dd) – itself an OCP effect. Thus the results demonstrate a 18.4 SOCIOLINGUISTIC CONTRIBUTIONS continuum: the phonology of English disfavours TO FORMAL THEORY sequences of segments to the extent that those seg- ments are similar; sequences of partially similar In formal linguistics one encounters many exam- segments are partially suppressed by means ples of principles that are originally proposed as of CSD, and sequences of completely similar

Table 18.3 Subject pronoun expression in Spanish: dialect differences in the effect of specificity of reference with second-person singular pronoun tú San Juan, Puerto Rico Madrid, Spain Per cent overt Factor weight n Per cent overt Factor weight n pronouns pronouns [+specific] 48% 0.51 145 40% 0.72 58 [−specific] 69% 0.72 188 19% 0.50 150 Source: Camerson (1993: 325)

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 8 33/22/2010/22/2010 6:11:276:11:27 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 9

Table 18.4 Preceding segment effect: an OCP analysis Preceding segment Deletion N Per cent Factor weight /t,d/ [+cor, −son, −cont] – (categorical absence, i.e. 1.00) /s,z,†,z/ [+cor, −son] 276 49 0.69 /p,b,k,g/ [−son, −cont] 136 37 0.69 /n/ [+cor, −cont] 337 46 0.73 /f,v/ [−son] 45 29 0.55 /l/ [+cor] 182 32 0.45 /m,N/ [−cont] 9 11 0.33 vowels – – (nearly categorical retention, i.e. 0.00) Source: Guy and Boberg (1994 corpus: 3 Philadelphians)

(i.e. identical) segments are completely sup- mental grammar. So an obvious question is, how pressed, banned in underlying forms, and avoided similar or different are the mental grammars of in derived forms by epenthesis (thus, bisyllabic individuals who speak what we informally call raided vs monosyllabic rained ). ‘the same language’? Given the diversity that Results of this sort have two-fold implications. sociolinguistic research documents, even among First, they confirm the theory; here they show that members of the same speech community, what OCP is an active principle in English. But, sec- elements of their mental grammars are diverse, ondly, they show the principle to have graded, allowing differentiation, and what are the same, probabilistic effects, not only categorical ones. In allowing them to communicate in a language that the Guy and Boberg data we see not an ‘obliga- they believe they share? tory’ principle, but rather a contour preference In the mainstream tradition the mental gram- whose force is inversely proportionate to the slope mar is an apparatus with categorical, invariant of the contour: flat contours are prohibited, gentle properties. Hence even a small alteration to its slopes are disfavoured and steep slopes are pro- internal structure, if it produces any difference in moted. This preference operates gradiently in output, would count as a discretely different gram- sociolinguistic variation, not as a ‘knockout’ mar. Linguistic diversity, therefore, would imply effect. Words with disfavoured consonant that speakers have grammars that differ in some sequences like act are not categorically deleted in component or set of components (features, rules, some social dialects or styles and retained in constraints, constraint rankings, etc.). Different others; rather, they are variable in all dialects, but speech styles would be handled the same way: to always show higher rates of deletion than those shift among different speech styles or registers, with more favourable contours, like old. The speakers would have to switch between discretely sociolinguistic findings thus test the formal model different grammars. By implication, styles and – the OCP – and in this case confirm it, but recast speakers would exhibit categorical differences: it as a stochastic generalization. speaker A would use some form in some context The OCP example shows a formal principle, that speaker B did not or could not use. conceived in the Neogrammarian tradition as dis- In the social world, it is clear that diversity is crete and invariant, being reformulated in the light orderly: socially proximal individuals are linguis- of sociolinguistic evidence as a probabilistic gen- tically proximal as well; family members speak eralization. What might the impact of sociolin- more alike than those who are unrelated; people in guistic evidence be on larger elements of formal the same community or dialect region are more theory, such as an entire grammar? Formal lin- alike than those who live farther away; and people guistics since Chomsky attributes a speaker’s of the same ethnicity, class or age group also tend capacity to use language to a mental grammar that to be more similar than those of other groups. codifies and encompasses linguistic knowledge. What does this scale of sociolinguistic similarity This is constructed by speakers in the course of correlate with in the mental grammar? Is the scale language acquisition, based on input from the a gradient continuum, or is it composed of dis- language they hear spoken around them (along crete steps, so speakers who are more dissimilar with, presumably, some universal elements). simply have a larger number of quantum differ- This model presupposes that the speaker is ences in their grammars? embedded in a social universe, with other speak- Sociolinguistic research suggests nuanced ers from whom the individual hears evidence answers to such questions. We encounter both needed for language acquisition, and with whom discrete and continuous differences between the individual communicates using his or her varieties, and we encounter different orders of

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 9 33/22/2010/22/2010 6:11:286:11:28 PMPM 10 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

quantitative diversity. First, much sociolinguistic demands of a situation, or for expressive value. variability is indeed continuous. One finding, by Social proximity between individuals is reflected now classic, is that speech communities are in their settings of this probability: one’s experi- replete with variables showing social stratifica- ences in language acquisition, daily interaction tion: higher-status speakers use more and lower- and accommodation, and attitudinal factors such status speakers use less of a given variant and, as solidarity, all combine to make it likely that simultaneously, all speakers use more of that vari- speakers who belong to common networks and ant in more careful styles. Labov’s (1966) findings who share social characteristics such as class, on the use of coda /r/ in New York City cogently ethnicity, etc., will have common values for these illustrate this pattern. probabilities. And what the speech community as The data in Figure 18.2 clearly suggest a con- a whole shares, including New Yorkers of differ- tinuum of usage; New Yorkers make finely graded ent social classes and ethnicities, is a common quantitative adjustments in their individual pro- social evaluation of the use of the /r/ variant: they duction, and are collectively finely graded in /r/ all understand it is associated with higher-status usage by socioeconomic status. It is implausible speakers and more formal styles, and reflect this to perceive in these data a series of discretely dif- shared evaluation in their own usage. ferent grammars, each associated with a particular However, beyond these scalar patterns of varia- speaker in a particular speech style, each display- tion associated with style, class, age, etc., we also ing some quantum difference from socially or encounter discrete differences. The data presented stylistically adjacent grammars. Instead, it sug- above in Tables 18.2 and 18.3 demonstrate this, gests a quantitative component of the mental lin- showing discrete differences between speech com- guistic capacity that controls the frequency of munities. Notice, however, that those examples coda /r/ – in other words, a probability. Each involved the effects of context on linguistic varia- speaker has a characteristic value for this proba- tion: the effect of a following pause on English bility, but they can adjust it continuously for sty- CSD, and the effect of specificity of reference on listic purposes, to accommodate the social pronoun expression in Spanish. Notice further that

SEC 80 6–8

9 60 4–5

2–3

1 40 0 /r/ index

20

0 ABCDD′ Style

Figure 18.2 Class stratification of coda /r/ in New York City. (from Labov, 1972: 114). Socioeconomic class (SEC) scale: 0–1, lower class; 2–5, working class; 6–8, lower middle class; 9, upper middle class. Style scale: A, casual speech: B, careful speech: C, reading style: D, word lists: D’, minimal pairs.

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1100 33/22/2010/22/2010 6:11:286:11:28 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 11

these discrete differences lay between different tested against reality and found wanting. Language speech communities – Philadelphia vs New York, is neither Platonic nor digital; it is a neurological, San Juan vs Madrid – whereas the continuously physiological and social system, an analog device varying data of Figure 18.2 come from speakers in that doesn’t always do the same thing in exactly the same community, in fact the same neighbour- the same way. An adequate formal theory must hood (the Lower East Side of New York). recognize inherent variability, and generate it. This suggests a basic distinction between two Secondly, the sociolinguistic evidence suggests dimensions of quantitative variability. We distin- that an adequate model must incorporate quantifi- guish overall rates of use of a variant (a context- cation. This point is still the subject of some free statement of the alternation) from differential debate, as some scholars propose to leave the rates of use by linguistic context (context- grammar unquantified and derive quantitative reg- sensitivity). Quantitative variation in the overall ularities observed in language use by other means, rate of use of social variables, differentiating as an epiphenomenal consequence of the range of classes, speech styles, etc., is in principle continu- possibilities (cf. Anttila’s OT analyses of variation ous, and speakers within a given community are – see below), or by some mechanism external to accustomed to, and accomplished at, varying the grammar proper that selects among different freely in this dimension. But the contextual effects, grammars. We shall argue, however, that these although they may also be probabilistic, do not alternatives are not adequate to account for the full vary freely within the grammar of an individual, range of quantitatively orderly heterogeneity. or within a community. Thus there is no social Thirdly, an adequate model must accommodate class of speakers in Philadelphia, for example, for social information. It requires representations that whom pause is a more favourable context for dele- model speakers’ capacities to speak in socially tion, and no speech style in which Philadelphians adaptive ways, and to perceive the social signifi- use pause as a favourable context. Hence we cance of the usage of others. Since individuals advance a hypothesis: the rankings, and quite pos- smoothly vary their usage both expressively and sibly the specific probabilities, associated with responsively, thereby communicating social mean- constraints on linguistic processes are a fixed fea- ings and adapting to social expectations, an ade- ture of the grammar of a speaker, and are one of quate formal model of this capacity must the shared linguistic characteristics common to all incorporate a social semantics not only of words members of a speech community. but also of linguistic forms and processes. The This distinction is given formal representation formal account must also be capable of represent- in the ‘variable rule’ model (Labov, 1969; ing diversity at the community level, differences Cedergren and Sankoff, 1974; Sankoff, 1978). in the usage of others that speech community This model associates an ‘input’ probability (p0) members can passively recognize the social sig- with every option or rule in the grammar; in addi- nificance of, even if their own grammars do not tion, context-sensitive processes may have proba- productively generate them. bilistic weights associated with relevant contexts The sociolinguistic adequacy of a formal model (contexts i, j, k, etc., are associated with weights can therefore be evaluated on the basis of these pi, pj, pk). In this model, the value of p0 is what three criteria: accounting for variation, quantifica- varies within a community, whether stylistically tion and social information. Let us consider some or socially, while the values of specific constraints formal models that have been proposed to address pi, pj, pk are fixed. This is an original contribution sociolinguistic issues, and how they fare by these of sociolinguistic research to the theory of gram- criteria. mars, specifically to the question of grammatical The first issue, of modelling variation, has similarity and difference. been widely addressed, by a variety of formal approaches. The variable rule (VR) model, men- tioned above, uses a generative framework in which probabilities are attached to operations; 18.5 TOWARDS AN ADEQUATE conventional obligatory processes have a proba- FORMAL LINGUISTICS bility of 1, but probabilities of less than 1 indicate variable outcomes. This is the model most widely What does an adequate theory of language need used in sociolinguistic research on variation. In – one that accounts for speakers’ capacity to pro- OT, several proposals have been made to generate duce and deal with diversity? The first require- variability using variable or underspecified con- ment is that it must accommodate variation. The straint rankings (Anttila, 1997, 2002; Nagy and Neogrammarian-inspired model of exceptionless Reynolds, 1997): instead of a fixed constraint linguistic processes has been mistaken in much of hierarchy that always gives the same outcome, formal linguistics for a design principle of lan- speakers are hypothesized to vary among several guage; in fact it is a hypothesis, which has been rankings of the relevant constraints which select

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1111 33/22/2010/22/2010 6:11:286:11:28 PMPM 12 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

different variants. In treatments of historical variation consisted of choosing one grammar or change, diachronic variability has been modelled the other for each relevant sentence they uttered. by means of alternation between grammars that What changed across time was the probability of generate old and new variants (Kroch, 1989b, speakers selecting one or the other grammar, 2000; Yang, 2000). All of these models are ade- while within each grammar, there was no variation. quate for generating variable outcomes. While providing an elegant treatment of his- Greater differences between approaches are torical change, this strategy does not offer a gen- found when we consider the second criterion, eral quantitative model for sociolinguistic quantification: How should a formal model variation. First, it does not, strictly speaking, solve account for the probabilistic patterns of natural the problem of variation; rather, it relocates it language? Do they result from one grammar or from inside the grammar to outside. To account from alternation between grammars? Should they for the quantitative facts of orderly heterogeneity be generated directly, by quantification within the – all the subtle probabilistic variation evident in linguistic system, or derivatively, as a secondary sociolinguistic usage – this approach requires an consequence of the way a grammar operates to additional theory of grammar selection. Secondly, generate a range of options, or the diversity of it runs afoul of Occam’s razor – the elementary representations or inputs? scientific principle of economy, that proliferation The quantificational strategy that involves the of explanatory elements should be avoided. For smallest departure from the theoretical tradition of the English do case, Kroch requires two grammars invariance is grammar competition. In this that are identical in every respect except for the approach, traditional invariant grammar is pre- V-to-INFL parameter setting. But at any given served, and variation and change are taken as moment, languages typically have not one but evidence that speakers command multiple gram- many variables – thus ME also has variable dele- mars, each one discretely generating one observed tion of that complementizers (I thought THAT he variant. These alternative grammars are selected knew vs I thought he knew), among other varia- with different probabilities to produce the quanti- bles. If each alternant of each variable requires a tative patterns that are found in language use. This separate grammar, and these variables intersect, so approach is exemplified in work on diachronic that a speaker may in one utterance use periphras- change by Kroch (1989b), Yang (2000) and others. tic do and complementizer deletion and, in the Kroch (1989a), for example, takes this approach next, periphrastic do but complementizer reten- to model the development of periphrastic do in tion, and so on, then a language with n binary English. In early Middle English (ME), before AD variables would require speakers to maintain n! (n 1300, negative constructions inserted not after a factorial) mental grammars, each trivially differ- verb or auxiliary (e.g., They know not what they ent from the next. Besides constituting a massive do), and questions were formed by inverting sub- violation of Occam’s razor, this would rapidly ject and verb (e.g. Slept you well?). In late ME and exhaust the storage capacity of the brain, if the early Modern English, modern equivalents with language had as few as a score of variables. auxiliary do began to appear (They don’t know The alternative to grammar competition is to what they’re doing; Did you sleep well?); for the step off the Neogrammarian path and locate vari- next 400 years, usage varied between the two ation inside the grammar. But there are still com- forms, with the modern constructions steadily peting strategies on how to formalize this so as to increasing in frequency. achieve quantitative adequacy. The VR model Kroch accounts for these facts in terms of a uses explicit quantification; in principle, it per- parametric shift. The grammar of early ME per- mits the probabilistic quantification of all gram- mitted the main verb to raise to an auxiliary (or matical operations and contextual conditions. INFL) position, so that it could do whatever aux- This allows it to achieve excellent accuracy in iliary verbs do (in Modern English other auxilia- accounting for the quantitative patterns observed ries also occur before not and invert with subjects in sociolinguistic variation. However, some schol- to form questions: e.g. They must not know; Have ars have argued that incorporating a probabilistic you slept well?). But in late ME an alternative component in the grammar gives an overly power- grammar emerged that did not permit verb-to- ful formal model, capable of generating patterns INFL raising, so that when the auxiliary position that do not occur. A notable alternative is found in was separated from the verb by another element, versions of OT using variably ordered constraints such as an inverted subject or a negative, do was (VOC), which treat quantitative patterns as, in inserted as an empty auxiliary to carry tense and a sense, epiphenomena – statistical reflections agreement. In Kroch’s analysis, during the four or of the range of possibilities permitted by gram- five centuries when this variation was evident, matical options, rather than the product of a spe- speakers entertained two mental grammars, one cific probabilistic element. This approach is with and one without V-to-INFL raising, and their exemplified by Anttila’s treatment of phonological

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1122 33/22/2010/22/2010 6:11:286:11:28 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 13

variables in several languages (Anttila, 1997; To illustrate this point, we can extend the previ- Anttila et al., 2008). ous example. Suppose, in another speech style, or As a simplified example, consider a VOC treat- a different group of speakers, a fourth constraint is ment of English CSD. Suppose there were only added to the variably ranked group, an OCP con- three relevant constraints: one disfavouring syl- straint prohibiting sequences of overly-similar labic codas (NoCoda), one disfavouring complex consonants, ruling out cases like act, mist, but codas (*CxCoda), and one favouring the faithful permitting left, old. When this constraint outranks articulation of underlying consonants (FAITH). If FAITH, deletion is selected in words like act. the ranking of these constraints was unspecified in Hence when OCP is variably ranked with respect the grammar, and randomly varied whenever to the other three, undeleted forms of act are speakers articulated a relevant word, they could selected only one-quarter of the time (the possible occur in any of six possible orders (NoCoda high- orders in which FAITH is highest ranked), rather est, followed by the other two in either order; than one-third, raising the expected deletion rate *CxCoda highest, followed by either order of the from 67 per cent to 75 per cent. But this constraint other two, and FAITH highest, with either order of has no effect on deletion of words like old, which the other two). Following standard OT conven- will continue to show deletion in 67 per cent of all tions, we assume the grammar selects outputs cases. Hence the consequence of involving an consistent with (i.e. incurring the fewest viola- additional variably-ranked constraint is to intro- tions of) the highest ranked constraints. Therefore, duce a new difference between contexts, not to when FAITH is highest-ranked, words like act, change deletion rates across the board. As we have mist are pronounced with final /t/, but when either noted, this is not how sociolinguistic variation of the other constraints outrank FAITH, a pronun- actually works; styles and social groups in the ciation with a deleted /t/ results, because NoCoda same speech community differ in overall rates of and *CxCoda are both violated by candidates use of variables, not in contextual constraints. preserving the final stop. Since two of the six pos- Finally, consider the third criterion, the formal sible orders generate retention of /t/, and the other modelling of social information. The above exam- four generate deletion, this model predicts dele- ple shows the difficulties OT encounters in this tion should occur in two-thirds of occurrences. respect; it could incorporate social information by Using the VOC strategy, Anttila has achieved associating styles or social indexicality with cer- impressively accurate quantitative models of sev- tain constraint rankings, but this incorrectly pre- eral cases of linguistic variation. He also argues dicts regular differences between speakers or that VOC explains dialect patterning – why cer- styles in contextual constraints on variables. VR tain combinations of variants are dialectally models treat social information as probabilistic attested, while other patterns we might imagine, constraints on production; thus stylistic con- but which are not generated by any possible order straints fit into the model in the same way as lin- of constraints, do not occur. These results present guistic constraints. In speech community studies a serious challenge to other formal models using VR, it is routine practice to treat social addressing these questions. However, the approach groups in the same way; so age or gender groups, is inadequate as a general model of sociolinguistic for example, are treated as if they were contexts variation. As we have noted, quantitative variation for usage associated with probabilities that has two aspects: differences in overall rates, increase or decrease the use of a variable. However, which vary between speakers and across speech this practice has a hazy theoretical status. It may styles, and differences in constraint rankings and more or less accurately model the collective contextual effects. VOC can model the latter, but behaviour of a community, but it is not clear what lacks any formal treatment of the former. The it means for the grammar or usage of an individual. only way predicted probabilities can differ in An important lacuna in all of these theoretical VOC, among styles, speakers or dialects, is by approaches to sociolinguistic variation is that, varying the number of constraints that are variably since they focus on variable production, they have ranked. In the simplified example above, three little to say about the perception and mental repre- constraints have 3! = 6 possible rankings; if four sentation of diversity, so that speakers’ passive constraints were involved, they would have 4!, or knowledge of the social significance of the usage 24 possible rankings. So adding more constraints of others is not explicitly accounted for. It is a to the set that are variably ranked gives more pos- central argument of sociolinguistics that speakers sible rankings, and therefore different possible understand the social significance of the linguistic proportions among the selected variants, but it behaviour of other members of their speech com- also means the variation will now occur or fail to munities, even when these others are doing things occur at different rates in whatever contexts are with their language that the speaker does not per- affected by the constraints that have been added to sonally do. This understanding is reflected in the mix. shared evaluations of variables, and it influences

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1133 33/22/2010/22/2010 6:11:286:11:28 PMPM 14 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

production in such ways as the shared directions also incorporate memories of who uttered them of style shifting. But how is this knowledge repre- and what the social circumstances of usage were. sented in the mind? Should it have formal repre- In this way, a speaker can deduce from memory sentation in the mental grammar? To conclude our what variants are more often associated with what consideration of formal models, we turn now to a social context or identity. This allows them to theory that puts such representations at the centre interpret the usage of others, and to adapt their of linguistic structure. own usage accordingly. An indexical association between linguistic characteristic and social char- acteristic emerges statistically from the distribu- tion of exemplars in the cloud. 18.6 NEW DIRECTIONS: USAGE-BASED Such a model addresses all the shortcomings of GRAMMAR conventional formalisms, providing a framework for explicitly modelling inherent variability, An important theoretical development that explic- orderly heterogeneity and social embedding. This itly seeks to offer a formal account of usage, vari- is an attractive framework for sociolinguistics. ation and change, and to provide a model of But there are open questions about the theory speakers’ knowledge that incorporates social and concerning its capacity to explain the things that stylistic information, is the emergence of usage- the theoretical mainstream was designed for and based approaches to grammar, notably Exemplar does well, such as categorical processes and Theory (Bybee, 2001; Pierrehumbert, 2001). This abstract operations. Many linguistic generaliza- theory is a radical departure from mainstream tions are indeed categorical: articles precede formal linguistics; it embraces variability and nouns in English, without exception; many sound rejects categorical, exceptionless processes. changes show Neogrammarian regularity: Indeed, it denies the existence of general abstract Germanic languages have no residual voiced aspi- processes, along with abstract representations of rates (/bh, dh, gh/) left over from Proto-Indo- words and segments. Instead, Exemplar Theory European in lexically-diffused defiance of (ET) begins by postulating that speakers remem- Grimm’s Law. Speakers use language in ways that ber, in rich phonetic detail, the tokens of words go beyond the evidence, for which they have no they hear pronounced, or produce themselves. exemplars. Therefore, speakers have stored in memory the For example, how does a child language learner full range of variation they have encountered, and do something new, which he or she has not use these memories (the ‘exemplar cloud’) as tar- encountered in the input? Consider the case of gets for their own productions, which necessarily Jesse, an American-English (AmEng)-speaking also vary. Words are not decomposed into strings child, observed by the author, who was moved to of phonemes, but remembered holistically; hence Australia at the age of 1;11. After two months of each word can behave distinctly, showing idiosyn- exposure to Australian English, he abruptly refor- cratic patterns of variation and change that diverge mulated his pronunciation of post-tonic intervo- from other words containing the same phonemes. calic obstruents: he made them all voiceless. The The substance of other approaches to phonology, only exemplars in his Australian English (AusEng) such as segments, morphemes and phonological input relevant to this change were AusEng pro- processes, are, in ET, all secondary generaliza- nunciations of [t] in words like water, little, which tions and analogic operations, rather than primary contrasted with the voiced flap in his native components of representation and grammar. AmEng phonology. But rather than changing just Exemplar theory thus incorporates a compre- the words for which he had such exemplars, Jesse hensive rejection of Neogrammarian exception- repronounced his entire vocabulary, affecting lessness, phonemic representation, categorical and other medial flaps, other voiced stops and even invariant grammar, and the langue/parole and voiced fricatives: thus daddy>datty, table>tapu, competence/performance distinctions. It puts vari- doggy>dockie, fuzzy>fussy. During this period he ation and diversity at the centre of the theory. It is acquired, from his AmEng-speaking parents, the inherently quantified: frequencies, proportions words driver and driving, and duly pronounced and contextual effects can be deduced from the them drifer and drifing. Aside from the intervo- distribution of variants in the exemplar cloud. But calic /t/ cases, he had no exemplars whatsoever perhaps its clearest advantage for sociolinguistics for any of these new pronunciations; in fact, is that, with modest elaboration, it can be extended the input from American parents, Australian peers to incorporate a formal representation of speakers’ and childcare workers offered massive counter- knowledge of the social significance of linguistic examples. Jesse’s behaviour suggests not a proba- variables, including their passive knowledge of bilistic exemplar-driven adjustment to his lexicon, the use of others. Thus Foulkes and Docherty but rather, an abstract reanalysis, a generalization (2006) suggest that the memories of exemplars of the form [+obs] → [-voice]/´ V___V.

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1144 33/22/2010/22/2010 6:11:286:11:28 PMPM SOCIOLINGUISTICS AND FORMAL LINGUISTICS 15

ET addresses such issues by recognizing proc- Anttila, A., Fong, V., Benus, S. and Nycz, J. (2008) ‘Variation esses of generalization and analogy, but these are and opacity in Singapore English consonant clusters’, vaguely defined and formalized. It is not clear Phonology, 25: 81–216. how the generalizations across exemplars and Bailey, C.-J. N. (1973) Variation and linguistic theory. experience that ET permits are substantively dis- Arlington, VA: Center for Applied Linguistics. tinct from the abstract grammar it rejects. Notably, Bayley, R. (1994) ‘Consonant cluster reduction in Tejano recent work in this framework takes steps to English’, Language Variation and Change, 6(3): 303–26. address such issues by formally recognizing Bresnan, J., Dingare, S. and Manning, C. (2001) ‘Soft con- abstract operations (cf. Pierrehumbert, 2006). straints mirror hard constraints: voice and person in English There are also empirical challenges to certain and Lummi’, in M. Butt and T. H. King (eds), Proceedings predictions of ET. For example, it is a central of the LFG 01 Conference, University of Hong Kong, online claim of the model that production is strongly proceedings. Stanford, CSLI Publications: http://csli- influenced by such statistical properties of the publications. stanford.edu. input as lexical frequency. This hypothesis is in Bybee, J. (2001) Phonology and language use. Cambridge: the process of being empirically tested, with Cambridge University Press. mixed results to date – some studies confirm sig- Callary, R. E. (1975) ‘Phonological change and the develop- nificant frequency effects on variation, while ment of an urban dialect in Illinois’, Language in Society, others are more equivocal. Guy et al. (2008), for 4(2): 155–69. example, found a frequency effect on CSD in Cameron, R. (1993) ‘Agreement, functional compensation, early New Zealand English – final coronal stops in and nonspecific tu in the Spanish of San Juan, Puerto Rico, words that are more common in everyday usage and Madrid, Spain’, Language Variation and Change, 5(3): were more likely to be deleted, but this result 305–34. depended on using frequency counts from the Cedergren, H. J. and Sankoff, D. (1974) ‘Variable rules: per- corpus under study. Standard frequency counts formance as a statistical reflection of competence’, from a large corpus of contemporary English (the Language, 50(2): 333–55. CELEX corpus) did not have a significant effect. Chomsky, N. (1964) Current issues in linguistic theory. The Hague: Mouton. Chomsky, N. (1965) Aspects of the theory of syntax. Cambridge, MA: MIT Press. 18.7 CONCLUSION Clark, B. (2005) ‘On stochastic grammar’, Language, 81(1): 207–17. For over a century the dominant formal theories in Clements, G. N. and Keyser, S. J. (1983) CV phonology: a linguistics have paid little attention to the findings generative theory of the syllable. Cambridge, MA: MIT or concerns of sociolinguistics, because they treated Press. the USE of language as fundamentally different Foulkes, P. and Docherty, G. (2006) ‘The social life of phonet- from, and not particularly relevant to, the mental ics and phonology’, Journal of Phonetics, 34: 409–38. linguistic capacity of human speakers, which was Guy, G. R. (1980) ‘Variation in the group and the individual: the declared concern of theoretical linguistics. In the case of final stop deletion’, in W. Labov (ed.), Locating place of natural language, they studied an idealiza- language in time and space. New York: Academic Press. tion. But the limits of this strategy are now becom- pp. 251–65. ing clear: the idealization cannot explain the real Guy, G. R. (1991) ‘Explanation in variable phonology: an world. If anything unifies the field of sociolinguis- exponential model of morphological constraints’, Language tics, it is a shared concern with the real world, in the Variation and Change, 3(1): 1–22. form of language in use. By conjoining the findings Guy, G. R. (1992) ‘Contextual conditioning in variable lexical and interests of sociolinguistics with the best prod- phonology’, Language Variation and Change, 3(2): ucts of formal linguistic thinking, we can illumi- 223–39. nate a path towards a genuine science of language. Guy, G. R. and Boberg, C. (1997) ‘Inherent variability and the obligatory contour principle’, Language Variation and Change, 9(2): 149–64. Guy, G. R. and Boyd, Sally (1990) ‘The development of a REFERENCES morphological class’, Language Variation and Change, 2(1): 1–18. Anttila, A. (1997) ‘Deriving variation from grammar’, in Guy, G. R., Hay, J. and Walker, A. (2008) ‘Phonological, lexical F. Hinskens, R. van Hout and L. Wetzels (eds), Variation, and frequency factors in coronal stop deletion in early New change and phonological theory. Amsterdam: John Zealand English’, poster presented at Laboratory Phonology Benjamins. pp. 35–68. 11, Wellington, New Zealand. Anttila, A. (2002) ‘Morphologically conditioned phonological Hymes, D. H. (1972) ‘On communicative competence’, in J. B. alternations’, Natural Language and Linguistic Theory, 20: Pride and J. Holmes (eds), Sociolinguistics. Harmondsworth: 1–42. Penguin.

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1155 33/22/2010/22/2010 6:11:286:11:28 PMPM 16 THE SAGE HANDBOOK OF SOCIOLINGUISTICS

Kiparsky, P. (1982a) Explanation in phonology. Dordrecht: Paul, H. (1978 [1886]) ‘On sound change’, in P. Baldi and R. Foris. N. Werth (eds), Readings in historical phonology: chapters Kiparsky, P. (1982b) ‘Lexical phonology and morphology’, in in the theory of sound change. Tr. by H. A. Strong from S. Yang (ed.), Linguistics in the morning calm. Seoul: Prinzipien der Sprachgeschichte. University Park: the Hanshin. pp. 3–91. Pennsylvania State University Press. pp. 3–22. Kroch, A. (1989a) ‘Function and grammar in the history of Pierrehumbert, J. (2001) ‘Exemplar dynamics: word frequency, English: periphrastic “do’’, in R. Fasold and D. Schiffrin lenition and contrast’, in J. Bybee and P. Hopper (eds), (eds), Language change and variation. Amsterdam: Frequency and the emergence of linguistic structure. Benjamins. pp. 133–72. Amsterdam: Benjamins. pp. 137–58. Kroch, A. (1989b) ‘Reflexes of grammar in patterns of lan- Pierrehumbert, J. (2006) ‘The next toolkit’, Journal of guage change’, Language Variation and Change, 1(3): Phonetics, 34(6): 516–30. 199–244. Poplack, S. (1980) ‘Sometimes I’ll start a sentence in Spanish Kroch, A. (2000) ‘Syntactic change’, in M. Baltin and C. Collins y termino en español: toward a typology of code- (eds), The handbook of contemporary syntactic theory. switching’, Linguistics, 18(7, 8): 581–618. Malden, MA: Blackwell. pp. 629–739. Sacks, H., Schegloff, E. A. and Jefferson, G. (1974) ‘A simplest Labov, W. (1966) The social stratification of English in New systematics for the organization of turn-taking for conver- York City. Washington, DC: Center for Applied Linguistics. sation’, Language, 50: 696–735. Labov, W. (1969) ‘Contraction, deletion, and inherent variabil- Sankoff, D. (1978) ‘Probability and linguistic variation’, ity of the English copula’, Language, 4: 715–62. Synthese, 37: 217–38. Labov, W. (1972) Sociolinguistic patterns. Philadelphia: Santa Ana, O. (1992) ‘Chicano evidence for the exponential University of Pennsylvania Press. hypothesis: a variable rule pervades lexical phonology’, Labov, W. and Fanshel, D. (1977) Therapeutic discourse: psy- Language Variation and Change, 4(3): 275–88. chotherapy as conversation. New York: Academic Press. Saussure, F. de (1916) Cours de linguistique générale. Labov, W. and Waletzky, J. (1967) ‘Narrative analysis’, in Published by C. Bally, E. Sechehaye, and A. Reidlinger. J. Helm (ed.), Essays on the verbal and visual arts. Seattle: Paris, Geneva: Payot. University of Washington Press. pp. 12–44. Verner, K. (1978 [1877]) ‘An exception to Grimm’s law’, in Leben, W. (1973) ‘Suprasegmental phonology’. PhD P. Baldi and R. N. Werth (eds), Readings in historical dissertation, Massachusetts Institute of Technology. phonology: chapters in the theory of sound change. Tr. Lim, L. T. and Guy, G. R. (2003) ‘The limits of linguistic com- by R. Stanley, from Eine Aushnahme der ersten munity: speech styles and variable constraint effects’, Penn Lautverschiebung. University Park: the Pennsylvania State Working Papers in Linguistics: Selected Papers from University Press. pp. 3–22. NWAVE 32, 10(2): 157–70. Weinreich, U., Labov, W. and Herzog, M. (1968) ‘Empirical Myers-Scotton, C. (1993) Social motivation for codeswitching: foundations for a theory of language change’, in W. P. evidence from Africa. Oxford: Clarendon Press. Lehmann and Y. Malkiel (eds), Directions for historical Nagy, N. and Reynolds, B. (1997) ‘Optimality Theory and linguistics. Austin, TX: University of Texas Press. variable word-final deletion in Faetar’, Language Variation pp. 95–195. and Change, 9(1): 37–55. Yang, C. (2000) ‘Internal and external forces in language Nishimura, M. (1997) Japanese–English Code-switching: change’, Language Variation and Change, 12(3), syntax and pragmatics. New York: Peter Lang. 231–50.

55434-Wodak434-Wodak CChap-18.inddhap-18.indd 1166 33/22/2010/22/2010 6:11:286:11:28 PMPM