Dialectology and Linguistic Diffusion.

Dialect variation brings together language synchrony and diachrony in a unique
way. Language change is typically initiated by a group of speakers in a particular
locale at a given point in time, spreading from that locus outward in
successive stages that reflect an apparent time depth in the spatial dispersion
of forms. Thus, there is a time dimension that is implied in the layered boundaries,
or isoglosses, that represent linguistic diffusion from a known point of
origin. Insofar as the synchronic dispersion patterns are reflexes of diachronic
change, the examination of synchronic points in a spatial continuum also may
open an important observational window into language change in progress.
In its ideal form, the spatial-temporal interaction may be displayed through
an appeal to a version of the wave model, in which a change originating at a
given locale at a particular point in time spreads from that point in successive
layers in a way likened to the waves in water that radiate from a central point
of contact.1 As a hypothetical example of the spatial-temporal reflex, let us
assume that there are three linguistic innovations, or rule changes, within a
language: R1, R2, and R3. We assume further that all three changes originate
at the same geographical location, the focal area for the language change. Each
one starts later temporally than the other, so R1 is the earliest innovation, R2
the next, and R3 the third (figure 24.1).
At time i, R1 is present at the location where the change originated but not
in outlying areas.2 At time ii, R1 may have spread to an outlying area while
another innovation, R2, may have been initiated in the focal area. At this
point, both R1 and R2 are present at the focal site, R1 alone is present in the
immediate outlying area, and neither R1 nor R2 may have spread to an area
further removed from the focal area. At time iii, the first change, R1, may
have spread to the more distant area, but not the later changes, R2 and R3.
In this hypothetical pattern of diffusion, we see that the successive dialect
areas marked by isoglosses – that is, lines delimiting the boundaries of each of
these rules – in geographical space reflect successive stages of language change
over time.

IMAGEN
Figure 24.1 The wave model of linguistic diffusion

The model represented in figure 24.1 is conceptually appealing, but it is also
simplistic and it often ends up begging essential descriptive and explanatory
questions about the empirically documented facts of dialect diffusion. What
are the social and linguistic mechanisms whereby forms spread, and what is
the transitional phase like? What kinds of diffusional configurations result
from the process? And, given that it has been maintained that the dialect
boundaries represented by isoglosses are “a convenient fiction existing in an
abstract moment in time” (Carver 1987: 13; and see our discussion of this point
below), what might an empirically motivated, dynamic model of diffusion
look like? To a large extent, our discussion will concern itself with establishing
the kinds of conditions and qualifications that need to be set on an ideal, abstract
model of diffusion in order to connect it with the empirical facts of dialect
distribution and to delimit the documentable patterns of diffusion. Our focus
is thus on the transition and embedding questions with respect to language change
rather than the actuation question, which addresses why language changes take
place to begin with (Weinreich et al. 1968).3
Although dialect diffusion is usually associated with linguistic innovations
among populations in geographical space, a horizontal dimension, it is essential
to recognize that diffusion may take place on the vertical axis of social space as
well. In fact, in most cases of diffusion, the vertical and horizontal dimensions
operate in tandem. Within a stratified population a change will typically be
initiated in a particular social class and spread to other classes in the population
from that point, even as the change spreads in geographical space. For
example, Labov’s research (Labov 1966, 1972a; Labov et al. 1972) indicates that
much change in American English is initiated in the working class and lower
middle class and spreads from that point to other classes.
We focus on the diffusion of dialect forms per se, but there is a fundamental
sense in which the transmission of linguistic innovation is framed by the broader
question of the diffusion of innovations. For example, Rogers (1983) argues that
there are at least five factors that influence the diffusion of customs, ideas, and
practices: (i) the phenomenon itself; (ii) communications networks; (iii) distance;
(iv) time; and (v) social structure. While linguistic structures present a
unique type of “phenomenon” for the examination of diffusion, the other factors
influencing diffusion, such as communications networks, distance, and social structure, are hardly unique to the dispersion of linguistic innovations. In fact,
our ensuing discussion should confirm the essential role of all of these factors in
linguistic diffusion, just as they figure prominently in other types of diffusion.
The framing of linguistic diffusion within a more general model of diffusion
however, should not be taken to mean that the social or “external” factors that
affect linguistic structure do so in ways that simply parallel their influence
with respect to other cultural phenomena. We maintain that there is a sense in
which the role of social factors in language change is fashioned to accommodate
the structure of language vis-à-vis other cultural phenomena. For example,
the current sociolinguistic position on the origin of change “universally points
to the working class and lower middle class as the originators of sound change
in contemporary American English” (Kroch 1978). This locus for the initiation
of change is quite different from that observed for other cultural phenomena.
With respect to technical advances, we know that middle-class groups, not
working-class groups, are the primary innovators of change so that primary
social diffusion comes from the top (Rogers 1983). For linguistic phenomena,
innovations initiated by the elite tend to be limited to borrowings from external
prestige groups (Guy 1988); members of higher social classes do not introduce
changes from within the language. The current sociolinguistic position on
the locus of change also differs from the traditional position within linguistics
(cf. Bloomfield 1933: 476; Joos 1952; Fisher 1958) that the lower classes strive to
emulate changes initiated by the upper classes in language as they do in other
cultural phenomena.
Furthermore, given the “natural” linguistic basis of many changes originating
in the vernacular speech of the working classes, it is convenient for a dominant
group to mark itself as linguistically distinct from the underclass by resisting
or inhibiting the changes toward “more natural” processes proffered by vernacular
dialect speakers. In such a model, natural linguistic changes spread
from the lower classes to the higher social classes when they are ratified
and evaluated as socially acceptable. The examination of linguistic dispersion
through a population may thus inform a more general model of diffusion
about the interaction of the innovative “phenomenon” and the social and
demographic factors that enable the process of diffusion.
1 Orderly Variation and Diffusion
All change necessarily involves variation.4 Speakers do not suddenly adopt
a new form as a categorical replacement for an older form, whether the form
involves a gradual, imperceptible change in the phonetic value of a vowel
within a continuum of phonetic space or an abrupt, readily perceptible change
involving the metathesis of consonants or the linear realignment of constituents
within a syntactic phrase. Instead, there is a period of variation and coexistence
between new and old forms in the process of change. This transitional period of fluctuation has often been ignored in historical linguistics under
the assumption that language change cannot be directly observed. Further,
variation in language has traditionally been dismissed as unsystematic and
irrelevant, a reflection of linguistic performance rather than competence and,
hence, of no bearing on models of language change or diffusion. However,
as Weinreich et al. put it: “The key to a rational conception of language change
– indeed, of language itself – is the possibility of describing orderly differentiation
in a language serving a community . . . in a language serving a community,
it is the absence of structured heterogeneity that would be dysfunctional”
(1968: 101).

TABLE
Table 24.1 Variation model of change.

An empirically based model of the dynamic process of diffusion must
recognize a variable transition period in the spread of dialect forms. However,
this transitional period is not one of chaotic, random fluctuation; instead, it
is a stage of systematic variability, or “ordered heterogeneity” that guides
language change meaningfully toward completion. Following Bailey (1973),
we hypothesize that there are a number of stages that change goes through in
the transition from the categorical use of one variant to its categorical replacement
by another. In between these two points are variable stages that show
systematic constraints sensitive to internal linguistic and external social factors.
Furthermore, the systematic variability of fluctuating forms will correlate
synchronic relations of “more” and “less” to diachronic relations of “earlier” or
“later” stages of the change. This is perhaps best shown by setting up a simple,
ideal model of the stages of change, as we do in table 24.1. Table 24.1 shows
the change from the categorical use of one form, X, to another, Y, in two
different linguistic environments, E1 and E2. Fluctuation between the forms is
indicated by X/Y.
Although the variable stages of change do not always follow the ideal
model for a number of reasons (cf. Bailey 1973; Fasold 1990; Wolfram and Schilling-Estes 1996), there is ample documentation in the quantitative sociolinguistic
literature (Labov 1980a, 1994) to affirm the general applicability of
this model of change to a broadly based set of language situations.
The notion of linguistic “environment” in such a model may be defined in
terms of a structural context, such as a syllable position in the case of phonological
change or a phrasal configuration in a syntactic one, or it may be
defined in terms of lexical sets. In other words, the model itself is impartial
to the Neogrammarian–lexical diffusion controversy that has underscored the
ongoing development of theories of phonological change over the last couple
of decades (see Labov 1981, 1994; Kiparsky 1988, 1995b (reprinted in this
volume); Hale, this volume). Furthermore, it should also be understood that
the notion of variability in this model applies to both intra-speaker and interspeaker
variation. In other words, an individual speaker will go through a
period of fluctuation between the old and new variant, and speakers within a
given speech community will show variation from speaker to speaker with
respect to the use of the new and old variant.
To illustrate, consider the case of h in English words such as hit [hIt] for it
[It] and hain’t [hent] for ain’t [ent].5 There is ample documentation (Pyles and
Algeo 1982; Jones 1989) that h was present in these words in earlier forms of
English and that it is still found to some extent in isolated regions of the
United States such as Appalachia, the Ozarks, and some Eastern coastal islands
(Wolfram and Christian 1976; Wolfram et al. 1997). At one point, h was found
invariantly in these items in both phrasally stressed syllables (e.g., Hít’s the one
I like) and unstressed syllables (e.g., I líke hit).6 The occurrence of the h in these
items then began to fluctuate (sometimes h occurred and sometimes not in
the production of a given speaker) in unstressed syllables while it was still
maintained invariantly in stressed syllables. Next, the h was variably deleted
in both unstressed and stressed syllables, but it was more frequently deleted
in unstressed syllables, where the change first started. Through time, the h was
completely lost in unstressed syllables while it was variably deleted in stressed
syllables. And finally, h was lost in both stressed and unstressed syllables
categorically. The stages of this change are summarized in table 24.2, using h
to indicate the categorical presence of h, h/Ø to indicate its variable presence,
and Ø to indicate categorical absence.
Among American English dialects today, stages 3 and 4 are still represented
in various isolated rural vernacular varieties and stage 5 is current mainstream
standard English usage, where the loss of h in it is complete.7 As found in this
example, the dialect differences in the use of initial h indicated among different
sets of speakers represent an ongoing change at different stages in its progression.
Although table 24.2 presents a simplified picture, given other social and
linguistic complexities involved in the distribution of this trait, it serves as a
model of the progressive steps that typically characterize the orderly dispersion
of a dialect form, as well as a model of a language change still in progress in
some dialect areas.
The lectal–temporal relation of tables 24.1 and 24.2 is necessarily based upon
the apparent time assumption, which has become a basic analytical construct within sociolinguistics over the past three decades (Labov 1963, 1994; Chambers
1995; Bailey et al. 1991). The fundamental assumption of the apparent
time construct is that, other things being equal (e.g., social class, dialect contact,
etc.), differences among generations of adults will mirror actual diachronic
developments in language (Bailey et al. 1991). From this perspective, the speech
of each generation is assumed to reflect the language as it existed at the time
when that generation learned the language. While the apparent time construct
has been applied almost exclusively to inter-generational differences within
the same speech-community, it seems appropriate to extend this construct to
the analysis of the geographical dispersion of language change as well (Bailey
et al. 1993). For example, we assume that h-dropping in hit and hain’t represented
in table 24.2 spread from the urban, focal areas of change in the United
States into outlying rural areas in successive stages. The change is complete in
these urban areas and therefore can no longer be observed “in progress.” At
the same time, the change can still be observed in progress in some more rural
areas, as successive generations of speakers exhibit stages 3, 4, and 5. 8.

TABLE 24.2
Table 24.2 Stages of change in the loss of h in (h)it and (h)ain’t in American English.

It is typically assumed in quantitative sociolinguistics that an increase or
decrease in the incidence of a particular linguistic variant in apparent time
indicates an expansion or recession of a change, respectively. Thus, we assume
in table 24.2 that the decreased use of the initial h in it and ain’t by younger
speakers in a given community is indicative of a change toward the loss of the
initial h. While this assumption matches the empirical facts in this instance,
it is not always the case that inter-generational differences reflect unilateral diachronic change. The most obvious exception to the apparent time assumption
is the phenomenon of age-grading, where the use of a form is associated
with a particular stage in the life cycle of a speaker. For example, teenagers
may use a particularized set of lexical items that are associated with this stage
of life; however, these items will be abandoned later in adult life because they
are no longer age-appropriate. Meanwhile, the next generation will proceed
through a similar cycle.
There are more subtle exceptions to the apparent time assumption, namely
reversals and pseudo-reversals of change in progress. These cases require more
elaborate cross-sectional analysis to determine and explain the pattern of change.
One of the best-known instances of a seeming reversal of a change in progress
is that presented by Labov (1963) in his analysis of the raising of the nuclei of
/ay/ and /aw/ in Martha’s Vineyard, an island located off the coast of Massachusetts
that has been a noted vacation spot for generations. Labov demonstrated
that on this island, while older residents showed a movement toward
the lowering of the traditionally raised nuclei of /ay/ and /aw/, middle-aged
speakers reversed this trend. This reversal is maintained to a somewhat lesser
extent by younger speakers, most likely as a way of asserting their islander
identity against mainlanders who flock to the island in ever-increasing numbers.
We have found a pattern of raising and backing for the nucleus of the /ay/
vowel on the Outer Banks island of Ocracoke, located off the coast of North
Carolina, which suggests, at first glance, that the recession of raised /ay/ is
being reversed in a way parallel to that reported by Labov (1963) for raised
/ay/ and /aw/ in Martha’s Vineyard (Wolfram and Schilling-Estes 1996).
Ocracoke Island, settled in the early 1700s, is located about 20 miles from
mainland North Carolina and, to this day, is not accessible by road. Thus,
Ocracokers were isolated not only geographically but socially for about two
and a half centuries. Shortly after World War II, the longstanding isolation of
Ocracoke Island came to an end, and a vibrant tourist industry transformed
the island. Historically, the Outer Banks region was characterized by a distinct
production of /ay/ which was close to the phonetic value [Oy] – a production
which has led to their characterization as “hoi toiders,” for high tiders. In
Ocracoke, as in Martha’s Vineyard, there are select groups of middle-aged
speakers with more /ay/-raising than older speakers. This patterning suggests
a reversal of a change in progress that parallels the Martha’s Vineyard case,
especially when coupled with the fact that the lowering of the nucleus of /ay/
to a low central vowel is a process which affected most varieties of English
at some point in history. However, when we compare the youngest group
of speakers with both the middle-aged and oldest generations of speakers
we find a dramatic decrease of [Oy] for the young speakers compared with
the two older age groups. Thus, the overall pattern of change across the three
generations does not show a reversal of a change in progress but a temporary
revitalization of the traditional variant before the complete erosion of [Oy].
The “pseudo-reversal” we observe in Ocracoke is, then, quite different from
the reversal of change reported for Martha’s Vineyard.
The orderly transition of linguistic forms not only shows systematic relationships
between the relative use of variants in terms of earlier and later stages of
spreading forms. Change also tends to show a characteristic trajectory slope in
the relative rate of progression through its transitional stage. Most variationists
(Weinreich et al. 1968; Bailey 1973; Labov 1994) maintain that there is a prototypical
rate of change which applies to the dispersion of new forms. This
pattern appears to apply both to the adoption of new forms on an individual
level (Bailey 1973) and to the spread of forms within a new community
(Weinreich et al. 1968). Change tends to start out at a slow rate, progressing
rapidly in mid-course, and then slowing down again in the last stages, modeling
the trajectory of an S-shaped curve. The change slope applies to change on an
intra-speaker and inter-speaker level; it also applies to change taking place
along a vertical or horizontal plane. As Bailey et al. note:
Like diffusion through the social spectrum, spatial diffusion takes place in a
three-part temporal process that simulates an S curve, with a period of infancy,
of slow expansion, during which the trait is relatively uncommon; a middle
period of rapid expansion after a critical threshold has been reached; and a later
period of saturation and filling in as potential adopters become scarce. (1993: 366).

Such a model has implications for several different dimensions of the diffusion
process, including the observation of diffusion in progress. For example,
the relatively rapid rate of progression through the mid-course of change makes
this period of change less accessible to direct observation than change at its
endpoints. The window for observing change in progress will be open longer
at the endpoints of the change trajectory – when the older or newer form is
clearly predominant – than at a midpoint of change when the fluctuation
between forms is likely to be most balanced between the use of the new and
old variant.
There are also implications about the orderly progression of change and the
role of the lexicon in change that seem related to the progression slope. For
example, the role of the lexicon in phonological change is more prominent at
the incipient and cessation stages of a change than at its midpoint. Furthermore,
we expect relationships of “more” and “less” in the relative use of variants to
be more directly correlated with “earlier” and “later” stages of the change during
the more rapid and maximally generalizable expansion period for new forms
than at the endpoints of the change. In fact, we submit that part of the resolution
of the ongoing controversy over regularity in phonological change may
be related to the trajectory slope of the change.9 From this perspective, irregularity
and lexical diffusion are maximized at the beginning and the end of the
slope and phonological regularity is maximized during the rapid expansion in
the application of the rule change during the mid-course of change. Our study
of the recession of the traditional production of /ay/ as [Oy] in Ocracoke and
the incipient diffusion of unglided [a:] from the Southern mainland bears this
out (Wolfram and Schilling-Estes forthcoming). The adoption of the Southern mainland unglided variant in Ocracoke is, at this point, still quite provisional,
being used in less than 10 percent of the cases where it might be used for /ay/.
And at this stage, it appears to be lexically constrained in its use and still
somewhat resistant to use in the most productive phonological environment
for ungliding indicated in the mainland South, namely, preceding a voiced
segment (Bernstein and Gregory 1994). We thus see that the progression slope
of change may be related to fundamental questions pertaining to the dynamic
process of change and diffusion.

2 Traditional Models of Linguistic Diffusion.

In our quest for a model of linguistic diffusion that fits the empirical reality
of change uncovered by sociolinguists and variationists, it is instructive to
review some traditional models for the spread of language change which
have been proposed within the course of dialectological and sociolinguistic
study. The traditional tree model used to illustrate the evolution of languages
has long been recognized by historical linguists and other language scholars
as inadequate for the description of diffusion (e.g., Hock 1991). This model
presupposes that closely related language varieties may suddenly sever all
contact with one another, diverging from that point into separate languages.
Linguistic innovations occurring in one of these language varieties after such
a split thus will be confined to that variety alone; the other varieties will be
left unaffected, even if they are found in the same geographic vicinity as the
innovating language. Of course, such a tree model is a highly idealized representation
of a far messier linguistic reality, since language varieties in close
proximity, even those genetically dissimilar, often influence one another in
profound ways.
To provide a conceptual picture for the area/rather than strictly genetic
spread of linguistic innovations, Schmidt (1872) developed the wave model
discussed above. To recapitulate briefly, the wave model holds that a given
linguistic innovation radiates outward from a central or focal area, in which
the change is usually carried through to completion. From there, the change
proceeds to a transitional area, in which the change occurs in varying degrees
of completion, depending on the distance from the focal point of change.
That is, a change which reaches an area adjacent to the focal area may occur in
almost all environments for the change, while one which is some distance
removed from this focal area may be effected in only one or two highly favored
environments. We have already seen how the loss of word-initial h in the pronoun
hit in American English spread in this manner from urban focal points.
At an early stage in the process of this change, the focal points for the change
would have been the locus for total or near-total loss of h in hit; surrounding
these urbanized focal areas would have been transitional zones in which the loss of h was incomplete in varying degrees. For example, in an area near the focal area, we may have found complete loss of h in unstressed position
and variable loss in stressed contexts; while in a transitional area farther
removed from the focal point, we might have found that h loss occurred only
variably, and then only in the most highly favored environment – unstressed
position. position.
It seems, then, that we could characterize the transitional area of traditional
wave-model-based approaches to linguistic diffusion, not as one dialect area,
but as a number of subtly different dialect areas. Trudgill (1983) refers to these
varieties as mixed lects – that is, dialectal varieties in which an innovative
variant alternates with a conservative variant. Trudgill (1983) also maintains
that there are so-called fudged lects – dialectal varieties in which the competition
between a new and an older form is resolved in favor of a compromise
form, perhaps a phonetic compromise, in the case of a sound change. For
example, Chambers and Trudgill (1980) note that in the transitional area
between the pronunciation of /u/ as [U] in the North of England and the
innovative [y] pronunciation that occurs in Southern England, we find both
types of intermediate language variety – mixed lects in which /u/ is pronounced
as [U] in some words and as [y] in others, and fudged lects, in which
/u/ takes on the phonetically intermediate value [ƒ]. From the perspective of
systematic variation we discussed earlier, however, a fudged lect seems to be
something of an anomaly, since speakers in an area on which an innovation is
encroaching typically show alternation between two variants – an older form
and the new form – rather than the sudden innovation of a third, compromise
variant. A “compromise” vowel is perhaps best viewed not as a resolution
between two competing vowels, but as a vowel which is currently located, as
it proceeds through a natural rotational pattern (e.g., Labov 1994), at an intermediate
point in phonetic space between a traditional vowel value and an
innovative pronunciation. For example, Trudgill’s “compromise” form [ƒ] is
most likely a synchronic reflex of the diachronic progression of [u] to [y], as
part of a vowel subsystem movement involving the lowering and unrounding
of high back vowels.10
Of course, as we have observed above, the very isoglosses we draw to
divide lectal areas from one another are, at least to some extent, “a convenient
fiction.” That is, dialect areas, particularly the subtly different areas which
comprise the transitional area for a linguistic change, cannot necessarily be
said to have clear-cut boundaries. The isogloss of one particular innovation
almost certainly will not correspond to that of another innovation, even if
each innovation radiates from the same focal point at the same point in time.
And even if we consider only one particular change when drawing our lectal
boundaries, the fact that innovative forms exist side by side with relic forms
throughout the transition area, in higher or lower proportion to the relic forms,
greatly hinders our efforts. For example, if in a given transitional area, a new
variant is used 80 percent of the time in one locale, 75 percent of the time in
another locale, and 70 percent of the time in a third region, are we to classify
each small area as its own lectal area, or should we perhaps draw a dividing line at, say, 75 percent usage of the new form? Girard and Larmouth (1993)
suggest that a transitional lect is perhaps best modeled not as a set of features
which either belong or do not belong to the lect in question but as a version of
a fuzzy set. In Girard and Larmouth’s interpretation, membership in the set is
characterized as a percentage figure rather than by a simple binary distinction
– that is, +/− Lect A. For example, suppose we wanted to describe a particular
lect, Lect A, as a fuzzy set composed of certain dialect features, including
loss of h in hit. If h loss in hit is variable in this dialect, we would not simply
say that initial h in hit is not a member of Lect A – that is, that the value of
h’s membership in Lect A = 0. Rather, we would express h’s membership in
the set of features comprising Lect A as a number between 0 and 1, say 0.75,
for example. Of course, exactly how such a figure might be determined is
a complex issue indeed and is beyond the scope of the present discussion.
The reader is referred to Girard and Larmouth (1993) for more discussion of
this point.11
Beyond the transitional area of a linguistic change we find what are traditionally
labeled relic areas – that is, areas which the innovation fails to reach.
Most often such areas are geographically distant from focal areas. Sometimes,
however, physical barriers to communication, such as mountainous terrain or
a body of water, may block the spread of a change from a relatively nearby
focal point. Social and demographic factors such as social and racial isolation
among neighboring groups may similarly play a significant role in delegating
areas to relic status. Thus, African American working-class groups in Northern
metropolitan areas within the United States may maintain some older
Southern rural dialect forms (e.g., the production of ask as aks or the use of
completive done, as in Kim done took out the trash) despite the fact that they are
one or two generations removed from their Southern roots. Patterns of racial
and social segregation have, in fact, greatly inhibited significant changes such
as the Northern Cities Vowel Shift (Labov 1991, 1994) from affecting inner-city
black communities, who remain immune to such changes while maintaining a
Southern-based vernacular dialect.
Areas which have been designated as relic areas with respect to one linguistic
innovation may very well be innovative, focal areas when another language
change is brought into focus (e.g., Hock 1991); thus, the designation of certain
areas as focal, transitional, or relic is largely relative, though demographic and
social factors such as population density may be favorable to the heavy concentration
of linguistic innovations in one particular area, such as a large,
centralized metropolitan area.

3 The Gravity Model.
A number of studies conducted by dialectologists and sociolinguists over the
past several decades indicate that the wave model is not empirically justified, even if it is expanded to incorporate systematic variability per our earlier
discussion. Trudgill (1974) demonstrated that a slightly different model, termed
the gravity model or the hierarchical model, provides a much better fit for the
observed data on dialect diffusion. According to this model, which is borrowed
from the physical sciences, the diffusion of innovations is a function not
only of the distance from one point to another, as with the wave model, but of
the population density of areas which stand to be affected by a nearby change.
Changes are most likely to begin in large, heavily populated cites which have
historically been cultural centers. From there, they radiate outward, but not in
a simple wave pattern. Rather, innovations first reach moderately sized cities,
which fall under the area of influence of some large, focal city, leaving nearby
sparsely populated areas unaffected. Gradually, innovations filter down from
more populous areas to those of lesser population, affecting rural areas last,
even if such areas are quite close to the original focal area of the change. The
spread of change thus can be likened not so much to the effects of dropping a
stone into a pond, as with the wave model, but, as Chambers (1993: 150) puts
it, to skipping a stone across a pond. Figure 24.2 illustrates such a model. Note
that larger circle sizes indicate increased population density.

IMAGEN
Figure 24.2 The gravity model of linguistic diffusion

The reason linguistic and other innovations often spread in a hierarchical
pattern is attributed to the fact that greater interpersonal contact is maintained
among places with larger populations, and heavy contact strongly promotes
the diffusion of innovations. This latter point was formulated by Bloomfield in
1933 as the principle of local density. However, even as the amount of interaction
between two areas is directly proportional to the population density of these
areas, so it varies inversely with the distance between the two locales – that is, interaction diminishes as the distance between two population centers increases.
This interplay between the population density of two areas and the distance
which separates them thus parallels the effects of density and distance on
gravitational pull (that is, the amount of influence two physical bodies exert
upon one another), according to the physical scientific gravity model.12
As early as the 1950s, geographers, most notably Hägerstrand (1952), used
the gravity model to describe and predict the diffusion of cultural innovations
such as technical advances. Trudgill (1974) adapted Hägerstrand’s model to
his sociolinguistic study of the Brunlanes peninsula of Norway to show that
the spread of a certain phonetic change – that is, the change in the phonetic
value of the /æ/ phoneme from [E] to [a] – was diffusing throughout the
peninsula according to the pattern predicted by the gravity model rather than
the wave model. The change to [a] began in a central area of relatively dense
population and proceeded from there to lesser centers of population and from
there to more rural, less populous areas. Further, Trudgill (1974) showed similar
patterns in the diffusion of linguistic innovations in the East Anglia area of
England. For example, he demonstrated that a generalized phonological process
of initial h-dropping currently underway in England (not the lexically restricted,
almost completed version we discussed earlier) spread from London directly
to Norwich, the population center of East Anglia, without affecting the thinly
populated area between the two cities. If the wave model accurately represented
the diffusion of linguistic innovation, the intervening area should have
been more h-less than Norwich, rather than almost completely h-ful, as it is in
reality.
A number of other studies reveal similar patterning whereby linguistic innovations
“skip” from one population center to another, leaving rural areas
unaffected until the final stages of the change. For example, Callary (1975)
showed that [æ] diphthongization in words like tag and bad and [a] fronting in
words like lock and pop spread from Chicago to downstate Illinois in a hierarchical
pattern; and Bailey et al. (1993) demonstrated a hierarchical pattern as
well for the diffusion of the /O/–/a/ merger in word pairs like hawk/hock
throughout Oklahoma. Even before the gravity model was applied to the study
of linguistic diffusion, the hierarchical spread of change was documented;
for example, Kloeke’s (1927) study of the change from [e] to [q] in medieval
Dutch and Flemish reveals a hierarchical rather than wavelike spread for this
change, while Kurath (1949) noted the importance of cities in the diffusion of
linguistic innovations along the east coast of the United States.
In most cases of hierarchical diffusion, the spread of innovation is from
relatively large regional centers to smaller, more localized towns and other
gathering-places. Occasionally, a change may reach a smaller city before a
slightly larger area, perhaps for geographic reasons, such as difficult terrain, or
for social and demographic reasons, such as a high concentration of a certain
social class in a given city. When changes actually do proceed strictly from
larger cities to smaller, we have so-called cascade diffusion. Such patterning has
been observed, for example, in the spread of musical influence from London in the 1960s throughout the world, first via other of the world’s major cities
and then gradually downward to small towns and rural areas (Haggett 1979).
Another type of diffusion that is recognized within the gravity model is
contagious diffusion. This refers to changes which actually do spread in wavelike
patterns – that is, changes whose spread is a primary function of distance only,
rather than population as well. Bailey et al. (1993) demonstrate such diffusion
in their investigation of the spread of a lax vowel [I] rather than tense vowel
nucleus [i] in the word field in Oklahoma speech. Interestingly, this contagious
diffusion co-occurs with the hierarchical diffusion observed for other innovations
in Oklahoma speech, as well as with yet a third pattern of diffusion,
contra-hierarchical diffusion, discussed below.

4 Limitations of the Gravity Model.

Although the gravity model accounts reasonably well for the diffusion of many
linguistic innovations, it falls short in a number of respects. As Trudgill (1983)
and Chambers (1993) point out, empirical evidence indicates the need for the
inclusion of factors other than distance and population into this model. For
example, the gravity model cannot account for the effects of terrestrial barriers
on the diffusion of innovation. And because the model was originally applied
to non-linguistic innovation, it cannot account for the unique effects of a linguistic
system itself on the spread of changes to that system. For example, the
gravity model cannot account for the effect of structural similarity in regional
language varieties with respect to the accommodation of new forms into
a region. Trudgill (1983) maintains that a dialect will more easily adopt an
innovation from a dialect to which it is highly similar than from a less similar
dialect. That is, “it appears to be psychologically and linguistically easiest
to adopt linguistic features from those dialects or accents that most closely
resemble one’s own, largely, we can assume, because the adjustments that
have to be made are smaller” (ibid.: 74). Trudgill attempts to correct for this
inadequacy in the gravity model by factoring into it a structural similarity
effect. That is, he describes the level of interaction between two centers not
only as directly proportional to the population density of each center and
inversely proportional to the distance intervening between them, but also as
directly proportional to how similar the two dialects are to one another.13
While Trudgill’s modification represents an important attempt to improve
the sometimes simplistic gravity model, it too has its limitations. As we have
discussed above, studies of variation within language and how this variation
leads to change indicate that the acceptance of a change depends on a number
of linguistic and social factors – not just mere similarity between dialects. For
example, in our study of the diffusion of external innovations into Ocracoke
English (e.g., Wolfram et al. 1997; Wolfram and Schilling-Estes 1996), we have
found that mere “similarity” does little to explain why certain innovations are adopted and others seemingly rejected. As Ocracokers lose their distinctive [Oy]
vowel, they may adopt one of two main innovating pronunciations: unglided
[a:], which is typical of mainland Southern speech, or the non-Southern variant
[aI], with a low central nucleus. We mentioned above that Ocracokers
are somewhat resistant to adopting Southern [a:], particularly before voiced
obstruents, the very environment where /ay/ is most likely to unglide in
mainland Southern speech. Yet in many respects, the Ocracoke dialect can be
said to be more “similar” to mainland Southern varieties than to non-Southern
dialects, particularly in its vowel system which, like the Southern vowel system
(e.g., Labov 1994), is characterized by the raising and tensing of the [I] and
[E] vowels (as in the pronunciation of fish as [fis]) as well as by the general
fronting of back vowels. Despite this similarity, there is quite a bit of resistance
in Ocracoke to the encroachment of the Southern [a:] variant vis-à-vis non-
Southern [aI]. Part of this resistance appears to be phonetic in nature: because
/ay/ most readily unglides to [a:] when its nucleus is a low central vowel [a],
Ocracoke [Oy] displays a degree of phonetic immunity to ungliding (Labov
1994). Further resistance to adopting the [a:] innovation stems from a complex
array of social factors, most notably the social meaning islanders attach to
the various /ay/ pronunciations they now come into contact with on a daily
basis. We return to this point below, in our discussion of contra-hierarchical
diffusion.

5 Amplifiers and Barriers to Diffusion.

As noted earlier, broad-based models of diffusion (Rogers 1983) encompass
at least five overarching factors that affect the spread of linguistic innovations:
the phenomenon itself, communication networks, distance, time, and social
structure. The gravity model takes into account the factors of distance and
communication networks (at least on a macro-level, as a function of population
density), while Trudgill’s model attempts to factor in a dimension related to
the diffusing phenomenon itself. Sociolinguistically grounded approaches to
language change tend to rely heavily on the role of social structures in the
diffusion of innovations.
Bailey et al. (1993) point out that just as topographical features may act as
barriers to the spread of innovation (or as amplifiers which encourage diffusion,
as in the case of a well-placed, easily navigable river), the social and
demographic characteristics of a region serve as even stronger barriers to and
amplifiers of change. Changes do not spread evenly across all segments of a
population, since some demographic groups are simply more resistant to or
accepting of change in general, or to certain specific changes, than others.
We have already mentioned that Labov’s research (e.g., Labov 1966; Labov
et al. 1972) indicates that members of “upwardly mobile” social classes, such
as the upper working class and lower middle class, as defined in traditional socioeconomic terms, are more quick to adopt innovations than members
of other classes. A number of other studies replicate these results (these are
summarized succinctly in Chambers 1995: ch. 2); and further studies show
that females are also among the leaders in linguistic change (see Chambers
1995: ch. 3 for a summary of a number of these studies). Further, younger
speakers are generally quicker to adopt new speech forms than older members
of a given speech-community.14 Thus it is instructive, in tracking the spread of
a change, to investigate the usage of a form not only across different regions,
but across different age groups and socioeconomic classes, as well as both
genders. For example, in their study of the spread of a number of linguistic
innovations across the state of Oklahoma, Bailey et al. (1993) demonstrate that
one can obtain a clear picture of the temporal spread of language changes by
examining the use of innovative forms in different age groups as well as in
communities of differing population densities.
Social factors other than age, gender, and social class also act as amplifiers
for and barriers to the diffusion of linguistic change, although their role is
not always as clear cut as is that of the above three factors. For example, the
presence of an ethnic minority population in an area may serve as an amplifier
for one particular change but yet act as a barrier in another area or with respect
to another change. Bailey et al. (1993) note that African American ethnicity
serves as an amplifier to r-lessness in Texas speech, while it acts as a barrier to
the spread of unglided /ay/ preceding voiceless consonants (e.g., right, like).
Barriers may further be classed as more or less permeable depending on how
strongly they block the spread of a change.
In addition, any thorough investigation of the effect of social factors on the
diffusion of linguistic innovations needs to include a closer look at communication
networks than that provided by the gravity model, which simply holds
that, in general, denser populations communicate more with one another than
do residents of sparsely populated areas. We find such a focus on micro-level
communication networks in the work of Lesley Milroy (e.g., Milroy 1980, 1987)
and James Milroy (e.g., Milroy 1992), who have done a number of studies on
the effects of the social networks of individual informants and small groups
of speakers on the diffusion of linguistic innovations (Milroy and Milroy 1985;
J. Milroy 1992; also see McMahon 1994a: 248–52 for an excellent summary of
the Milroys’ work on diffusion-related issues). The results of the Milroys’ social
network studies show that, in general, a population whose social networks
are dense and multiplex – that is, whose social networks involve frequent,
prolonged contact with only a small peer group, in a number of social contexts
– are more resistant to linguistic innovations than are populations whose
social ties are looser – that is, whose communications are spread out among
many people of different social groups and are, hence, briefer and less frequent
with each individual communicant.
Under the social network model for the spread of linguistic innovations, the
first people to adopt changes are those with loose ties to many social groups
but strong ties to none, since strong ties inhibit the spread of change. In order for the changes adopted by these people, called innovators, to make their way
into more close-knit groups, they need to be picked up by so-called early
adaptors – people who are central figures in tightly knit groups but who are
risky enough to adopt change anyway, perhaps for reasons of prestige (whether
overt or covert). Because these early adaptors are well regarded in their social
groups, the changes they adopt are likely to be picked up by other members of
these groups, thereby diffusing through a large segment of a population.
Given that urban populations are generally considered to be bound by looser
ties than rural societies, one can easily see how the Milroys’ model for linguistic
diffusion parallels the gravity model; both models maintain that innovations
begin in urban populations. The chief difference in the two models is that,
under the gravity model, increased interaction of any type leads to increased
diffusion of innovations; the Milroys maintain, however, that the interaction
must be of a certain type in order for innovation to spread. Further, the Milroys’
model affirms Labov’s conclusions that “upwardly mobile” social classes are
the quickest to spread innovations, for it is the individuals who make up these
classes who are most likely to maintain loose social ties with a number of people
from outside their immediate peer groups, as they strive to move out of their
current social class. Similarly, it is not surprising under this model that women
often lead linguistic change, since, in close-knit communities, it is usually
women who hold jobs that bring them into contact with members of social
groups other than their own (e.g., Chambers 1995: ch. 3).
Thus, the social network model may be seen not so much as a further description
of how linguistic innovations diffuse through a given population, but
rather as a potential explanation for the diffusional patterns that dialectologists
and sociolinguists have already observed.

6 Contra-Hierarchical Diffusion.

The final type of linguistic diffusion we examine points out just how strongly
social factors may affect the spread of language change. In their study of
the social and demographic factors that influence the diffusion of change in
Oklahoma speech, Bailey et al. (1993) discovered that, while one important
change, the spread of the /O/–/a/ merger, appeared to be diffusing according
to the hierarchical model, the spread of the quasi-modal fixin’ to (as in They’re
fixin’ to go now) displayed exactly the opposite diffusional pattern. That is, fixin’
to initially was most heavily concentrated in the rural areas of the state and
spread from there through increasingly large population centers until it reached
the state’s most urban areas. The different patterns of diffusion for these two
linguistic changes are given in figures 24.3 and 24.4, taken from Bailey et al.
(1993). In each illustration, figure (a) gives the spatial distribution of the form for
respondents born in or before 1945 and figure (b) gives the spatial distribution
for respondents born after 1945 in order to show the spread of the change in apparent time. Note that the circles on the maps represent cities, with larger cities
being represented as larger circles. Figure 24.3 illustrates how the /O/–/a/
merger, spread, in general, from more populous areas in Oklahoma to more
rural areas from the pre-World War II years to the present day. Figure 24.4
shows that fixin’ to spread in the opposite way, proceeding from rural areas to
larger cities over the course of the past several decades..

IMAGE
a Respondents born in or before 1945

IMAGE
b Respondents born in or after 1946

IMAGE
Figure 24.3 Spatial distribution of /a/ in hawk
Source: Bailey et al. (1993: 369)

IMAGE
a Respondents born in or before 1945

IMAGE
b Respondents born in or after 1946

IMAGE
Figure 24.4 Spatial distribution of fixin’ to
Source: Bailey et al. (1993: 372–3)

Given that communication, and hence the spread of innovations, is generally
held to increase with increased population density, how can we explain
the contra-hierarchical diffusion displayed by fixin’ to? Bailey et al. (1993) note
that there is an important difference in the social setting that contextualizes
the /O/–/a/ merger and that which contextualizes fixin’ to. While the identical
pronunciation of word pairs such as hawk/hock or Dawn/Don is generally held
by Oklahoma residents to be a mark of increasing urbanization or sophistication
and reaches its highest concentration in the speech of recent transplants to
Southern states such as Oklahoma, fixin’ to is regarded as a traditional, rural form
and is most prominent in the speech of Oklahoma residents whose Southern
heritage is well established. While many older Southern forms have indeed
faded in the face of newer Northern forms as more and more non-Southerners
migrate to the South, a number of rural forms have flourished, spreading from
outlying areas to population centers as long-time Southerners seek to assert their
identity against newcomers from the North. Bailey et al. maintain that fixin’ to
is one such form, as evidenced not only by the diffusional patterns observed
for this item but also by the strong correlation they found between informants
who use the form and nativity within Oklahoma, as well as positive valuation
of the state as a good place to live. Brown (1991) has also reported a contrahierarchical
spread for the merger of /I/ and /E/ before nasals (that is, the
pin–pen merger) in Tennessee.
In our study of Ocracoke English (e.g., Wolfram et al. 1997; Wolfram and
Schilling-Estes 1996), we have observed that the traditional pronunciation of
the /ay/ diphthong as [Oy] in Ocracoke serves as a marker of island identity,
much as does raised /ay/ in Martha’s Vineyard (Labov 1963). While [Oy] is
not diffusing beyond the confines of the island (or beyond the Outer Banks
island chain), its value as a marker of in-group identity does allow it to serve
as a barrier which blocks the incursion of non-Southern [aI] and especially
mainland Southern [a:] into Ocracoke speech. In other words, [Oy] is not spreading
contra-hierarchically as are some other traditional, rural forms of speech in
the American South, but it is not simply being supplanted by forms which are
moving down from major population centers, thanks in large part to its social
meaning in the Ocracoke community.
We have seen, then, that the social valuation accorded to linguistic forms
can drastically affect the process of linguistic diffusion. Linguistic markers of
local identity may serve as barriers to urban forms diffusing down into rural
areas; or these markers may be of such importance over a widespread region
that they actually take root and spread, effectively reversing the usual direction
of linguistic diffusion.

Explaining the empirical facts of dialect diffusion obviously calls for a
multidimensional approach that considers an array of geographical, social,
and linguistic factors which may interact in different ways. Furthermore, a
dynamic model of diffusion must encompass the systematic variability that
characterizes language change. While such a perspective exposes the inadequacies
of many of the traditional models for describing and explaining dialect
diffusion and some of the proposed alternatives, in the long run, a model
which remains continually sensitive to the emerging empirical facts about
dialect diffusion is the only one that will ultimately serve dialectologists and
linguists, whether their concentration is variation study or another area of
specialization. In the final analysis, our goal is to understand why and how
language changes over time and space.

ACKNOWLEDGMENT

NOTES:
1 Although the wave model of
language change goes back to the
nineteenth century (Schmidt 1872),
a revised model which incorporates
structured intra-language variability
into the model is used here. For the
most part, this model follows Bailey
(1973).
2 Weinreich et al. (1968) and Labov
(1994: 311) maintain that there is
no precise distinction to be made
between the origin of language change
and diffusion, since it is not the act of
innovation that changes language, but
the act of influence that instantiates
it. Thus, “the change and the first
diffusion of the change occur at the
same time” (Labov 1994: 311).
3 This is not meant to minimize
the significance of the actuation
question, which must ultimately be
considered in an authentic account
geared toward explanation.
4 Whereas all change necessarily
implies a period of variation, the
converse is not necessarily true.
That is, not all variation implies
change. Some variation may be
very stable and a product of the
natural performance of language
rather than an indication of
dynamic directionality with
respect to the replacement of
forms.
5 We are much more certain about
the earlier, widespread presence
of initial h in hit (Jones 1989: 245ff)
than we are about the h in ain’t. The
different sources which apparently
merged in the development of ain’t,
including haven’t, aren’t, and amn’t
(Cheshire 1982), and the socially
charged status of this item over
time have clouded the picture of
change somewhat. It is possible that
the h in ain’t was originally found only in those items derived from
haven’t and that this h-initial form
was subsequently generalized to
encompass tokens of ain’t derived
from amn’t and aren’t.
6 The differential distribution of
h loss in stressed and unstressed
syllables is still manifested in
h-initial pronouns in present
English. Note, for example the
difference between Him, I like
versus I like ’im.
7 Other varieties of English
(Trudgill 1990: 42), as well as
other Indo-European languages,
have a much more extensive
version of h-dropping that is
strictly phonologically conditioned.
In a sense, the lexically restricted
version illustrated here vis-à-vis the
generally applicable phonological
deletion process found in other
varieties of English (Jones 1989:
245ff) realistically illustrates a
case of the “regularity controversy”
with respect to dialect patterning
(Labov 1994).
8 We appeal to the uniformitarian
principle in assuming that the
patterning of current changes in
progress reflects the way in which
completed changes were effected.
This principle, as stated by Christy
(1983: ix), is that “knowledge of
processes that operated in the
past can be inferred by observing
ongoing processes in the present.”
See now the discussion in Janda
(2001: section 8) and in the
introduction to this volume.
9 The behavior of forms through the
course of the change slope is not
unlike the behavior of forms as they
proceed through the stages of early
and second language acquisition.
In language acquisition, a period
of rote learning of forms paves
the way for the application of an
exceptionless rule, which is then
realigned to accommodate the
empirical reality of regularity and
irregularity within language.
10 Our perspective here assumes
that vowel changes follow orderly
rotational schemes as set forth
in Labov (1994). The three main
principles guiding vowel rotation
are as follows:
Principle I: In chain shifts, tense
nuclei rise along a peripheral track.
Principle II: In chain shifts, lax
nuclei fall along a non-peripheral
track.
Principle III: Tense vowels move to
the front along peripheral paths, and
lax vowels move to the back along
non-peripheral paths.
In this schema, Trudgill’s fudged
lects would simply be a stage of
principle II (Labov 1994: 176).
11 Bailey’s (1973) framework is also
reminiscent of the fuzzy set model
for defining dialects, particularly in
transitional areas. Bailey maintains
that speakers of a language are
inherently polylectal – that is,
each of them possesses internal
grammars for a number of different
lects, which Bailey sets up on a
scalar, implicational array. In this
schema, the existence of certain
features will imply others, and
different lects are simply subsets
of the overall set of implicational
relations.
12 Even the mathematical formula
initially used to express the cultural
influence of two population centers
on one another is reminiscent of
physical scientific models for
gravitational effects. This formula is
as follows: Mij = PiPj/(dij)2, where
M = interaction, P = population, and
D = distance (e.g., Trudgill 1983: 74).
Trudgill (1974) first applied this
formula to the spread of linguistic innovations specifically, as opposed
to cultural innovations in general;
but, as we shall see, he expressed
serious reservations about its
adequacy as an accurate model
of linguistic influence.
13 In formulaic notation, the
revised model looks like this:
Mij = s · (PiPj/(dij)2), where s is a
variable expressing linguistic
similarity.
14 In fact, it is this sharp contrast in
younger versus older speakers’
adaptability to linguistic innovation
that allows sociolinguists to make
the apparent time assumption,
discussed at length above, on which
so much of their work is based.