Variation and Language, an Overview

Text-only Preview

Variation and Language, an Overview


Encyclopedia of Language and Linguistics 2nd Edition

Walt Wolfram
North Carolina State University
English Department, Box 8105
Raleigh, NC 27695-8105

The past half-century has witnessed remarkable growth in the study of language variation, and it
has now become a highly productive subfield of research in sociolinguistics. This section
considers the locus of variability within language and discusses the linguistic variable as a
heuristic construct in the analysis and the description of language variation. It further discusses
the nature of systematic language variation, as well the role of structured heterogeneity in
language change. Finally, it considers possible claims about representing language capability
with respect to language variation and implications for a grammar of language.


If structure is at the heart of language, then variation defines its soul. As Sapir (1921: 147) put it,
“Everyone knows that language is variable.” Furthermore, variation allows us to differentiate
individuals, groups, communities, states and nations. Notwithstanding the pervasive nature of
variability in language, it has often been disregarded or dismissed as tangential to the description
of structural patterning and irrelevant to the study of linguistic competence. In fact, it was not
until the advent of sociolinguistics a half-century ago that the admission of language variation
became more than a footnote to linguistic description. The study of language variation is now
one of the most rapidly expanding subfields of linguistics with a well-established cohort of
researchers, regular conferences, and scholarly journals, but its status is still somewhat marginal
within theoretical linguistics, notwithstanding the insistence of William Labov that the study of
language variation is central to the solution of fundamental problems in linguistic theory (e.g.
Labov 1966, 2001).

Variability is everywhere in language, from the unique details in each production of a
sound or sign to the auditory or visual processing of the linguistic signal. In fact, one of the
amazing facts about human communication is the demonstrated ability to normalize the inherent
variation within every spoken or signed message in processing the linguistic signal. Though
language variation is persistent and pervasive, it is not all equally interesting, even to those who
focus on the systematic nature of language variation, so-called LANGUAGE VARIATIONISTS. For
example, there is considerable variation in speech or sign production related to the physical
make-up of an individual speaker or signer—differences in vocal tract size in spoken language or
differences in the size of the hands and body used in signing—that is of interest to those who
study language normalization, but this is not generally the focus of systematic language variation
studies. Similarly, the relative fluency of production related to idiosyncratic behavior is not of

concern to those interested in language variation, though the dichotomy between meaningful
social differences and socially insignificant personal differences is not always clear-cut.
Generally speaking, interest in language variation focuses on differences that have some social
significance in terms of group behavior rather than personal idiosyncrasies, though socially
meaningful aspects of individual speaker performance are of interest to those interested in
language variation.
The empirical reality is that the boundaries of significant and insignificant language
variation are often gradient and obscure rather than discrete and transparent. There is, for
example, a fine line between consequential and inconsequential fluctuation in the vowel
formants of repeated productions of a given vowel in a constant phonetic context. Some
variability in each vowel production is expected, but a larger envelope of fluctuation in the
formants might be indicative of a significant vowel change in progress. Determining significance
for an envelope of vowel formants, however, can only be arrived at through a series of
exploratory procedures that takes into account individual speaker traits, social group norms, the
production of other vowels in the system, and so forth. Obviously, the empirical boundaries of
investigation for language variation studies are not as straightforward as they may appear in
descriptions of systematic variability found in the research literature and often involve
preliminary decisions about what to include and what to exclude in the study of orderly variation.
In traditional linguistic description the notion of variation within structural units has often
been acknowledged under labels such as “free fluctuation,” “optional rules,” and “free variants.”
Though long recognized, this variability has nonetheless been considered to be “an area of little
importance” (Crystal 2003: 189), a kind of garbage heap for variants that could not be predicted
invariably within the categorical framework assumed under most models of linguistic
description. But in keeping with the adage that “one person’s garbage is another person’s
treasure,” the examination of such variation has become the cornerstone of sociolinguistics, with
several professional journals dealing regularly with issues of language variation, including one
dedicated exclusively to this issue (Language Variation and Change), and prominent
sociolinguistics conferences that routinely feature presentations on language variation. This
includes perhaps the most influential sociolinguistics conference in the world, the annual
conference New Ways of Analyzing Variation (NWAV), now well into its third decade of
continuous growth.

An essential construct in the study of linguistic variation is the LINGUISTIC VARIABLE, a structural
unit that includes a set of fluctuating variants showing meaningful co-variation with an
independent set of variables. The linguistic variable was explicitly set forth in early variation
studies by William Labov (1966), the acknowledged founder of the field of language variation
studies, though this construct was certainly implicit in work previous to that point. On one level,
the relationship of the variable to its variants may be likened to that of the classic emic-etic
relationship in linguistic description, such as that between a morpheme and its allomorphs or a
phoneme and its allophones; however, as we shall see, it is hardly confined by such structural
relationships and boundaries. The unit is an abstraction since linguistic variables have no
concrete reality apart from the particular variants through which they are realized.
Operationally, the linguistic variable has been used to encompass a wide range of
fluctuating variants. The set may be a structural type, such as grammatical category (e.g. a tense
category, possessive category), a phoneme (e.g. the phoneme /θ/ or the phoneme /ŋ/ in English),

or a natural class of items in a specific linguistic context (e.g. coronal stops in syllable-coda
position). It may also be defined in terms of the application of a general process, such as the use
of a particular type of contraction (e.g. English variation between She has not vs. She hasn’t vs.
She’s not) or in terms of a syntactic relationship, such as concord (e.g. negative concord in
English as in He didn’t do anything about any problem vs. He didn’t do nothing about no
) or phrasal constituency (e.g. head-initial vs. head-final phrases). Variants are usually
established apart from theoretical models of language description and may, for all intents and
purposes, be accommodated within any syntactic model (e.g. Principles and Parameters,
Minimalism, Head-Driven Phrase Structure Grammar, Lexical Functional Grammar, and
Construction Grammar) or phonological model (e.g. Generative Phonology, Lexical Phonology,
Optimality Theory, Exemplar Theory), though there are significant implications for how such
variation may be accommodated within a particular model of language. Finally, the linguistic
variable may be defined as a simple lexical choice (e.g. soda, pop, tonic, or co(ca)-cola for a
carbonated drink in various regions of the US) or even in terms of a speech act (e.g. alternatives
for a expressing a ‘directive’ or a ‘compliment’).
As noted, the linguistic variable can cover a full range of linguistic units and
relationships. By the same token, the relationship of the variants to each other is not necessarily
limited by the fundamental typology operative in linguistic analysis and description. For
example, a given phonological variable may encompass allophonic fluctuation within a given
phoneme, different phonemes, or combine allophonic and phonemically distinct variants within a
unitary variable. Thus, the pronunciation of the vowel of English words like coffee or caught
may involve variants that range from different phonetic productions within the same phonemic
vowel (e.g. [kaofi] vs. [ko´fi]) to a phonemic merger with another phonemic unit (e.g. the vowel
of caught and cot are both pronounced as [Å]). In this case, the variants appear to be united by
the fact that they exist within a single lexical item. Furthermore, in some applications of the
linguistic variable, variants may involve competing systems or grammars (Kroch 1989) so that
variation cannot even be considered inherent within a unitary system. Under this interpretation,
typological correspondence or equivalency in the competing systems is the apparent basis for
including different variants in a linguistic variable.
The notion of the linguistic variable has been applied to a full range of levels within
language, though not without some argument about its appropriate application. Phonological and
morphosyntactic variation have tended to dominate language variation research in synchronic
and apparent time studies while syntactic variation has been a significant locus of investigation
in diachronic studies. For synchronic and apparent time studies, which have tended to use natural
conversational interviews as their primary databases, there is both a practical and theoretical
explanation for the focus on phonology and morphosyntax. In natural conversation, syntactic
phenomena simply do not occur at sufficient frequency levels for meaningful quantitative
analyses of systematic variability, the procedural underpinning for variation studies. But there
have also been theoretical concerns about the application of variation studies beyond phonology
related to the assumption of “semantic equivalency”, that is, the notion that there is no change in
referential meaning based on the selection of one variant or another (Lavandera 1978). While it
may safely be assumed that the variant pronunciations of unstressed –ing in the English word
running as a final velar [ŋ] or coronal nasal [n] have the same denotational referent, questions
about semantic equivalency have arisen in examining variation on other levels. For example, it
may be argued that there are subtle meaning differences in the Montreal French use of variants
être and avoir in different contexts (Sankoff & Thibault 1977). Similarly, it may be questioned

whether the difference between the ‘active’, ‘be passive’, and ‘get passive’ in English (e.g. They
broke into the store
vs. The store was broken into vs. The store got broken into) are authentic
alternatives for “saying the same thing” (Weiner & Labov 1983). Lavandera (1978) has
suggested that the condition of strict semantic equivalency need not be applied rigorously so that
the linguistic variable could be used in study language variation beyond phonology, a guideline
now generally adopted in studies of syntactic variables.
The focus on syntax in diachronic studies of language variation is, to a large extent, a
product of the kinds of available texts for investigation. Historical documents tend to be more
reliable for examining grammatical than phonetic and phonological variation. Furthermore, the
increasing accessibility of mega-corpora with automated search capabilities for historical texts
has made the investigation of syntactic variability much more accessible to variation researchers.
Though there are lingering questions about the precise structural status of the linguistic
variable, it remains a primitive construct in the study of language variation. Most of the lingering
questions and qualifications have to do with its precise status as a structural unit and its possible
relationship to a grammar of language rather than its usefulness as convenient heuristic in the
investigation of language variation. Whether or not it is explicitly recognized, the linguistic
variable remains at the procedural foundation of variation analysis, since one must assume that
the analyst can delimit a unified, exclusive set of fluctuating language variants that shows
meaningful patterns of co-variation with an independent set of variables. Though sensitive to
traditional structural categories and relationships in language, the linguistic variable is not
necessarily beholden to them, and there are clearly cases where the imposition of a traditional
linguistic boundary may actually detract from rather than enhance the understanding of language
variation in its social context (Wolfram 1993). Such considerations may be one of the reasons
why it might be argued that the linguistic variable is a uniquely sociolinguistic construct that
conveniently—if not always comfortably—covers the locus of co-variation between linguistic
variants and other variables that include social, historical, psychological, and linguistic factors.

As noted, the significance of the linguistic variable lies in its co-variation with other factors, or
independent variables. These factors do not typically correlate with the categorical use of a
variant but with the relative frequency of variant occurrence. For example, in most varieties of
English, there is fluctuation in the production of the final segment of words such as swimming
and walking with a coronal nasal [n] vs. a velar nasal [ŋ], commonly represented in spelling as –
in’ vs. –ing, respectively. Studies of the fluctuation between these variants over a half century
now (e.g. Fischer 1958, Shuy, Wolfram, & Riley 1967; Kiesling 1996) have shown that the
relative frequency of the in’ variant correlates with various types of social, psychological, and
linguistic factors. For example, Fischer (1958) showed t/hat boys tended to use more -in’ than
girls and that differences also related to speaker personality type (aggressive vs. cooperative) and
mood (tense vs. relaxed); Shuy et al. (1967) showed that –in’ was more frequently used by lower
status groups than higher status ones, and Kiesling (1996) showed that a group of fraternity men
used -in’ more frequently to project different types of masculine images during fraternity
meetings associated with relations of power and solidarity. At the same time, studies have shown
that there is variation in the use of the –in’ variation based on the lexical category of the word, as
verbs are more likely to occur with in’ than nouns (Fischer 1958). In all of these cases, the
difference in -in’ usage is a matter of relative frequency rather than categorical predictability—a

tendency in which variants have a greater or lesser likelihood of occurring under certain
Factors that correlate with higher and lower frequency levels of a given variant are
referred to as CONSTRAINTS ON VARIABILITY, where the term “constraint” is used to refer to a
factor that systematically correlates with increased likelihood that a given variant will occur.
This use is somewhat different from how the term is used in theoretical linguistics, where
constraint is used to refer to a condition that characterizes a universal principle of language or a
condition that restricts a language-specific rule. Factors that correlate with the increased
frequency of a variant are said to FAVOR the occurrence of a variant whereas those that correlate
with reduced frequency DISFAVOR or INHIBIT the occurrence of the variant.
Independent variables that co-vary with systematic differences in the relative frequency
of a variant are of two primary types, structural linguistic factors related to the linguistic system
itself, so-called INTERNAL CONSTRAINTS, and social or sociopsychological factors of various
types that exist apart from the linguistic system, so-called EXTERNAL CONSTRAINTS. Independent
linguistic constraints seem to align closely with traditional types of structural units considered
relevant in linguistic description. Thus, for systematic phonological variation, the feature
composition of the variant (e.g. voicing, sonorancy), phonetic environment (e.g. preceding and
following segments, stress patterns), hierarchical status (e.g. syllable position), and grammatical
status (e.g. type of morpheme) may be factors that constrain variability. There may also be other
factors, such as the lexical condition that high-frequency words favor a variable process over
low-frequency words (Myers & Guy 1997). For morphological and syntactic variables, lexical
category (e.g. noun vs. verb), phrasal composition (e.g. NP vs. VP, heavy vs. light phrases), co-
occurrence relations (e.g. concord, phrasal complements), embedding (matrix vs. embedded
clause), and adjacency conditions (proximate vs. distal) may be relevant factors affecting the
relative usage of fluctuating variants. These are, of course, the kinds factors considered in
traditional qualitative syntactic descriptions. As with phonological variables, this typology may
be extended somewhat, so that factors such as the serialization of structures in a narrative
sequence, for example, may affect variability along with the reified, sentence-level grammatical
categories and constituents generally recognized in syntactic description.

The types of external variables that correlate with the relative frequency of fluctuating
variants may include traditional demographic variables (e.g. age, social class, region),
constructed social groupings and practices of various types (e.g. communities of practice, social
networks), interactional dynamics (e.g. power relations, solidarity), and even personal
presentation styles and registers (e.g. performance, mimicking). Although there seems to be
somewhat of a disjunction between systemically based linguistic factors and a broadly ranging
array of social and psychological factors, most studies of systematic variation include both
independent linguistic and social factors within the same description of linguistic variability.
Furthermore, one of the principles that has guided variation analysis over the past few decades is
the “principle of multiple causes” (Bailey 2002:118), which holds that no single contextual factor
can satisfactorily describe the variability observed in natural language. Thus, linguistic factors
such as structural composition and linguistic environment may combine with a set of social
factors such as age, status, situational context, and so forth in the description of systematic

Though an assortment of linguistic and social variables need to be considered in
accounting for systematic variability, not all are of equal magnitude in their orderly effect on the
occurrence of variants. To sort out the relative influence of different constraints on variability,

variationists often use a type of multivariate regression analysis model designed by the
mathematician and linguist, David Sankoff, known as VARBRUL (Cedergren & Sankoff 1974).
Although a variety of commercially available multivariate regression analyses might be applied
to determine the significance of various factors on the co-variation of linguistics variants,
VARBRUL was designed specifically to handle the kinds of data found in language variation,
where not all of the logical cross-products may be manifested and there is often quite unequal
distribution of tokens in different cross-product cells in the sample.
VARBRUL is a probabilistic-based statistical procedure that shows the relative
contributions of various factor groups to the overall variability of items. The weighting values
range from 0 to 1 so that a value of greater than 0.5 in a binomial application indicates that the
factor being considered has a favoring effect on the occurrence of the variant, while a value of
less than 0.5 indicates a disfavoring effect, whereas in a trinomial application, a value of more
than 0.33 would indicate a favoring effect and a value less than 0.33 would be a disfavoring
effect. The higher the VARBRUL weighting is, the stronger the effect of the factor is in
accounting for the variability. Both Microsoft and MacIntosh versions of this program are readily
available and commonly used by researchers in language variation studies. In table 1, taken from
Mallinson & Wolfram (2002), results from a VARBRUL analysis of the incidence of syllable-
coda, stop-final cluster reduction in a remote Appalachian region of North Carolina (USA) are
presented, following a conventional procedure for reporting such data. In this binomial analysis
based on the application or non-application of cluster reduction, the input probability represents
the overall application mean of cluster reduction apart from the constraining effects. Two
independent linguistic variables, or FACTOR GROUPS, are considered in this simplified
presentation, the morphemic status of the cluster (e.g. monomophemic as in guest or mist vs.
bimorphemic as in guessed or missed) and the following phonetic environment (e.g. consonant as
in guest parking, vowel as in guest appearance, and pause as in guest //), and one social factor,
ethnic group membership (European American vs. African American). The total Chi square and
the Chi square score per cell (i.e. preconsonantal, monomophemic, African American/
prevocalic, bimorphemic, European American/ etc.) indicate the goodness of fit for the
quantitative data.

Table 1. Illustration of results of VARBRUL analysis (from Mallinson & Wolfram 2002)

Input probability = .29


African Americans = .65; European Americans = .33
Grammatical Status

Monomorphemic = .56; bimorphemic = .39
Following Environment

consonant = .80; pause = .37; vowel = .24

Total Chi square = 3.085 Chi sq. per cell = .257

The data indicate that a following consonant (.84) strongly favors cluster reduction over a
following pause (.37) and vowel (.24), and that cluster reduction is favored for monomorphemic
(.56) over bimorphemic (.39) clusters. Though both following phonetic environment and
morphemic status are significant constraints on variability in this study, the following phonetic

environment is a stronger constraint than the morphemic status of the cluster. The analysis also
reveals that ethnic group membership is a strong social effect on variability, with African
Americans (.65) favoring cluster reduction over European Americans (.35).

Though VARBRUL has become the regular statistical program for determining the
systematic effect of various constraints on variability, it is not without limitations. Some of these
are a product of the procedural assumptions but some arise from deeper questions about the
nature of variation itself. For example, one of the assumptions of VARBRUL is the
independence of factor groups, for example, the assumption that the effect of following phonetic
environment operates independently from the morphemic status of the cluster. As Sigley (2003)
points out, however, both linguistic and social factors are often interactive. With respect to social
factors, for example, ethnic group membership may work in tandem with the social factors of
age and social class rather than operate independently; similarly, for linguistic factors a
suprasegmental effect such as stress may interact with a segmental effect related to the following
or preceding phonetic environment. The assumption of independence in effects is therefore
questionable, leading to the development of more refined statistical procedures to tease out these
interaction effects (Sigley 2003). There are also lingering questions about how such linguistic
and social constraints might be represented in a grammar that includes systematic variability.

Another issue raised by procedures such as VARBRUL relates to group and individual
variation. VABRUL analyses are typically based on aggregate data, that is, pooled data for a
designated group of speakers without reference to an individual speaker. This procedure
obscures variation patterns for individuals, raising the issue of whether or not we can assume that
the individual and the group are the same with respect to systematic variation—a kind of
homogeneity assumption. Empirical investigations of individual speakers in relation to group
norms show that there is impressive regularity from speaker to speaker with respect to the
constraint effects on variation, once a reasonable number of tokens are extracted from individual
speakers. This leads Guy (1980: 12) to conclude that group norms “recapitulate the generally
uniform norms of individuals.” Some of the replicability of constraint effects for individual
speakers may, however, be due to systemic principles inherent within language itself. It makes
phonetic sense, for example, that syllable-coda cluster reduction would be much more likely take
place when the following segment is a consonant as opposed to a vowel, given the phonetic
complexity of consonant sequencing without intervening vowels. Other constraints, however,
seem language specific, such as the relationship between a following pause and following vowel
in their effect on variability (Labov 1994). Though there is sometimes an impressive replicability
in terms of constraint effects from speaker to speaker, there may be considerable variation in the
overall levels of variation for individual speakers, and some differences in relationships between
constraint effects for individual speakers within a group.
Detailed studies of groups and individuals show that one cannot assume that they are
always the same with respect to variation, and that there may be considerable variability among
speakers within a relatively homogeneous social group. In fact, Dorian (1994) suggests that there
is a kind of “personal-pattern variation” that does not co-vary with external social factors (Dorian
1994). Speakers are both individuals with idiosyncratic life histories as well as affiliated
members of a complex array of social groups, making it impossible to explain away all
individual-based variation in terms of group norms.

One of the most fruitful areas for the application of language variation analysis is language
change, given the fact that change necessarily involves variation. Speakers do not go to bed one
night using a particular form only to wake up the next morning to find the form categorically
replaced by another one. Variation exists whether the change involves a gradual, imperceptible
shift in the phonetic value of a vowel within a continuum of phonetic space or is an abrupt,
readily transparent change involving a major syntactic realignment of phrasal constituents. In the
progression of language change, there is a transitional period of co-variation between old and
new variants. Though all change seems to involve variation, this does not mean that all variation
necessarily implies change. Some variation may be stable, a product of internal systemic and
natural performance phenomena rather a reflection of dynamic directionality. Sorting out
dynamic and stable variation is, however, not always obvious, and can not necessarily be
determined a priori. In most cases, decisions about transitional and stable variation can only be
made after examining patterns of variation across time and by considering underlying
psycholinguistic and sociolinguistic principles that might help determine the difference.
Although the variability inherent in change has sometimes been ignored in historical
linguistics under the traditional assumption that language change could not be directly observed,
the empirical analysis of language change at different points in real time and in “apparent time”
(i.e. the assumption that differences across different generations of speakers at a given point in
time will mirror actual diachronic change) indicates that the variable transition of variants
proceeds in an orderly fashion over time and space.
Various models have been proposed for capturing the systematic heterogeneity associated
with change. One of the earlier models (Bailey 1973) proposed that change started variably in a
limited, natural linguistic environment and then spread from that point to other environments.
Furthermore, this model assumed that that variable change in the earliest environment would
ideally show greater variability than changes in other environments where the change started
later, so that relationships of “more” and “less” with respect to new forms might reflect
chronological relations of “earlier” and “later.” This relationship can be illustrated by setting up
a simplified, ideal model of the stages of change as in table 2, following Bailey (1973). The table
models the progression from the categorical use of one form, A, to another, B, in two different
linguistic environments, E1 and E2.

Table 2 Variation model of change following Bailey (1973)

Stage of Change

Categorical status, before undergoing change

Early stage, begins variability in restricted environment

Change in full progress, greater use of new variant in E1 where
change first initiated

Change progresses towards completion with categorically on new
variant first in E1 where change initiated

Completed change, new variant categorical

The model in table 2 illustrates the “sequential actuation” of change (Pintzuk 2003: 513), where
a change starts in one linguistic environment, a more favored environment, and then spreads to a
less favored one. However, speakers may also initiate a new form at the same time in different
environments, so-called “simultaneous actuation”, or initiate change at different frequency levels
in different environments. Cases of simultaneous change, along with other factors that may
accelerate or impede a change over time, call into question the assumption that “more” and
“less” can be interpreted without qualification as “earlier” and “later.”
The progression of variable change over time has now been subjected to considerable
scrutiny in historical linguistics, particularly with reference to syntactic change (e.g. Kroch 1989;
Kroch & Taylor 2000; Pintzuk 2003). Though it was originally suggested (Bailey 1973) that
more favorable contexts for change might progress at an accelerated rates by comparison with
less favorable contexts, research by Kroch and his colleagues supports the conclusion that there
is a Constant Rate Effect (CRE) in change, in which the rate of change for each context is the
same even though the frequency of competing forms may vary across linguistic contexts. It is
important to understand that the label “constant” does not refer to a constant rate of increase in
frequency over time, but to the fact that change will be the same across different linguistic
environments, so that the progression of change in E1, E2, and En will follow the same trajectory

The study of language change also indicates that there is a prototypical progression slope
for variation in time, where change typically follows that of an S-shaped curve (Bailey 1973;
Labov 1994; Pintzuk 2003). That is, the replacement of form A with B occurs slowly at the
beginning of a change, accelerates rapidly during the mid-course, and then tails off slowly in the
final stages of the change, as illustrated in figure 1.
Figure 1. Model of S curve in language change

This model has implications on several different levels. For example, the relatively rapid rate of
progression through the mid-course of change sometimes makes this period less accessible to
direct observation than the slower periods at the beginning and end points. It may also have
implications for the regularity of a change. For example, the role of lexical conditioning in
phonological change may be more prominent at the incipient and cessation stages of the change,
with lexical diffusion playing a more prominent role at the endpoints. Furthermore, irregularity
in constraint effects on variability may be more prone to take place at the beginning and end
points of the change than they might be during the mid-course of a change.


The examination of systematic language variation over time offers a unique opportunity
to observe language change in progress. More importantly, it has moved historical linguistics
away from a long-held belief that diachronic linguistics was limited to post-hoc analysis of the
end-products of language change (Weinreich, Labov, & Herzog 1968). Indeed, language change
has proven to be one of the most productive venues for studying the nature of systematic
variability as well as providing an important proving ground for examining the empirical validity
of different models of language change.

As a methodological heuristic, there is little question about the usefulness of variation analysis.
Its implications for the representation of speakers’ language capabilities and the role of variation
in a language grammar, however, is much more disputable. In fact, most theoretical linguists
would simply dismiss systematic variability as an artifact of performance with little relevance for
a model of language competence. Notwithstanding a few exceptions (e.g. Guy 2003; Anttila
2003), far too little attention has been paid to the role of systematic variability in a grammar of

Though the question of speaker knowledge and language variation has been raised on
occasion (Cedergren & Sankoff 1974), surprisingly little discussion has focused on the
capabilities that speakers might have with respect to language variation as a putative attribute of
language competence. Possible claims about speaker capabilities may include the following,
ranging from the weakest to the strongest claim.
• Speakers can identity optional (variable) variants
• Speakers can identify the factors that favor and inhibit variable application
• Speakers can identify the relative strength of different constraints on variable application
• Speakers can identify a probabilistic mechanism that generates differential constraint
• Speakers can identify frequency levels of variation
There is little psycholinguistic experimentation to support any of these claims, though there is
evidentiary documentation that the speaker-hearer can at least recognize optional variants and
distinctions between groups and individual speakers based on relative frequency levels. Insight
into the finer details of speaker knowledge, such as evidence for knowledge about the relative
strength of constraints, the recognition of a probabilistic mechanism for generating constraint
effects, and knowledge of frequency levels, however, still awaits psycholinguistic

In the early days of variation studies, researchers (Labov 1969; Cedergren & Sankoff
1974) proposed that variation was inherent within a language system and part of speaker
competence, and hence should be integrated into the grammar. This was expressed formally in
so-called VARIABLE RULES, which incorporated ordered constraint effects, and even probabilities,
into the formal generative-style rewrite rules in the tradition of Chomsky and Halle (1968). As
formal grammars shifted towards the formulation of universal principles rather than specific
language rules, however, the variable rule was abandoned (Fasold 1991). At the same time,
reformulated models of grammar raised new issues about the assumed inherency of variation that
was once considered to be the cornerstone of variation theory. In the Principles and Parameters
model of syntax, for example, it is assumed that parameters are set for a given language in one
way or another, while in Optimality Theory in phonology variability is reduced to different
rankings among the universal constraints on phonological structure. Accordingly, variation