Network analysis considers an individual's network, the
linkages, the strength of linkages and frequency of interaction, and
their impact on that individual. A variety of researchers from
anthropology (Barnes, 1954; Mitchell, 1969; Radcliffe-Brown,
1952), social psychology (Moreno,
1953), family sociology (Bott, 1957),
and health/stress management (Caplan,
1974) have found great explanatory power in individual-level
social network analysis. Boissevain pointed out that social networks
are viewed in two ways: (1) as a system of relations that impinge on
individuals and influence their behavior, and (2) as a series of
relations that persons use to achieve their ends (1973:viii). Sociolinguists have found social network
analysis to be a critical tool in explaining some linguistic behavior (Bortoni-Ricardo, 1985; Eckert,
2000; Gal, 1979; Labov, 1972; Milroy,
1980; Wolfram, 1974), very often with
reference to statement (1). Lesley Milroy's 1980 study of three working-class neighborhoods in
Belfast has been very influential in the use of social network analysis
in sociolinguistics and serves as the model for this study. In her
study, Milroy showed that strong ties to a social group can serve as
pressure to conform to that group's linguistic norms.
Milroy's study examined linguistic behavior and correlated that
linguistic behavior with a respondent's social network, showing
how the language use of members of these three Belfast speech
communities reflected patterns of personal interaction within the
communities. She concluded that “personal network structure is in
these communities of very great importance in predicting language use:
a dense, multiplex personal network structure predicts relative
closeness to vernacular norms” (1980:160).
Milroy's analysis considers social network as a composition of
two dimensions, density and plexity, that exist on a continuum. Thus,
an individual's social network may be described as more or less
dense/loose and more or less multiplex/uniplex. Density relates
to the structure of an individual's contacts. The more members of
an individual's social network who know each other, the more
his/her social network is said to be dense. Plexity refers to the
relationship to members within an individual's social network. If
many members are connected to the individual by more than one type of
relationship (e.g., a coworker is also a neighbor or a relative) then
that individual's social network is said to be multiplex. She
showed that there is a relationship between network structure and
linguistic choice and that this relationship can explain individual
linguistic behavior. For example, she found that the consistent
variation between two female respondents could not be explained by
social status, sex, age, or neighborhood (they were the same for both
respondents), but rather by the social network structure of the
individuals (1980:131).
PROFILES OF THE TWO DIALECTS IN CONTACT IN
YPSILANTI
The Northern Cities Shift
The Northern Cities Shift is a widely documented (Callary, 1975; Eckert,
1989; Fasold, 1969; Gordon, 1997; Ito, 1999;
Labov, 1994) ongoing sound change in urban
areas in the northeastern part of the United States, including Buffalo,
Cleveland, Detroit, and Chicago. It is slowly making its way to the
surrounding less-urban areas (Gordon, 1997;
Ito, 1999). The earliest description of the
NCS was given in 1969 by Fasold, who described the raised position of
,
the fronted position of /a/, and fronted and lowered position of
of respondents from Detroit.
These “new positions” for vowels are described relative
to an older vowel system of American English. Figure
1 shows such a system, based in part on Peterson and Barney
(1952). Although their analysis of American
vowels is based on a less than ideal sample of 76 speakers, most from
Mid-Atlantic areas, it has been usefully employed in studies similar to
the present one and has been reproduced with similar results by Stevens
and House (1963) and Hillenbrand et al.
(1995), the latter for Great Lakes area
speakers. Thus, the terms “raised,” “raising,”
“fronted,” “fronting,” and so forth, used here
and throughout refer to the relative position of the F1 and F2
characteristics of a vowel with regard to older documented vowel
systems (Figure 1) and not to tongue height
or movement. In addition, the raised quality of
refers to its onset. It is not the case that
becomes a high-front vowel throughout its duration; in fact, it diphthongizes,
and the remainder of the altered vowel reveals that it is a
“centering” diphthong.
A pre-Northern Cities Shift vowel system (Peterson & Barney,
1952, with additional /ey/ and
/ow/ data from Stevens,
1998).
Labov (1994) proposed that the NCS is a
change that progresses in several interconnected steps. Synchronic and
diachronic research (such as Eckert, 1986;
Gordon, 1997; Ito,
1999; Labov, 1994; and Labov, Yeager, & Steiner, 1972) has shown that
the fronted and raised position of
is present either before other elements of the NCS or very early in the
overall changes involved. Labov hypothesized that this shift in the location
of
creates a void in the vowel space that leads to the
fronting of /a/ and subsequent lowering and fronting of
.
Figure 2 (arrows pointing from the old position to
the new position) shows all the elements of the NCS. The status of the NCS as
a chain shift or related series of changes has been debated (Gordon, 2000; Stockwell & Minkova,
1997), but this debate goes beyond the scope of this article.
The Northern Cities Shift (adapted from Labov,
1994:191).
If speakers demonstrate the later stages of the NCS
(
backing and
lowering/backing), they are considered to be at an “advanced”
stage in the NCS. Speakers with advanced systems are usually female, young,
European American, and upper-working-class or lower-middle-class (Labov, 1994:156).
The NCS is a dialect feature of the Ypsilanti area. If the
respondents in this study have been affected by the local dialect,
there should be, at least, evidence of the oldest aspect of the NCS:
the fronting and raising of
.
Therefore, only that feature is considered here. This, of course, leaves open
the question of whether or not other features of the NCS or, indeed, other
features of Northern speech different from their own dialect are being
acquired by these Ypsilanti respondents. In this work, a finely-tuned analysis
of the acquisition of one feature is attempted. Preliminary analyses of
the data acquired for this study suggest that, in fact, only the first
two stages, the raising and fronting of the onset of
and the fronting of /a/, as characterized in previous research
on the shift, are present at all in the speech of these respondents.
Limitation to
was, therefore, practical as well as hypothetical.
The Southern Shift
The community investigated here, Ypsilanti, Michigan, is
distinguished by its large population of migrants from the Appalachian
region of the United States. The local vowel systems for the
respondents in this study, who were born in the South, is certainly
different than that of Southeastern Michigan. The Southern United
States is also involved in an ongoing sound change, but, in many cases,
with the opposite results. In the Southern Shift (Feagin, 1986; Labov, 1991, 1994; Labov, Yeager, & Steiner, 1972) shown in Figure 3 (again with arrows pointing from the old
position to the new one),
is fronting (but not raising, as in the NCS),
are not moving, the onset of /ey/ is found in the
position (or lower), and
diphthongizes with its onset moved to the /ey/ position. This reversal
is also found with /iy/ and /I/.
The Southern Vowel Shift (adapted from Labov, Ash,
& Boberg, 1997).
Acquisition of raised
,
rather than a general or even specific loss of Southern Shift features, is
considered in this study.
The Ypsilanti community
Ypsilanti is located in Washtenaw county, Michigan, approximately 30
miles west of downtown Detroit and 10 miles east of Ann Arbor.
Ypsilanti Township, on the eastern side of Ypsilanti, was the site for
the Willow Run bomber plant, built by Ford Motor Company in 1941. Ford
Motor Company recruitment programs and word-of-mouth drew many
southerners to work in the plant in the early part of the 1940's.
Of the 207,000 white Americans who migrated to the Detroit–Willow
Run region between 1940 and 1944, 83,930 came from the states of
Kentucky, Tennessee, Alabama, Mississippi, Ohio, Indiana, Illinois, and
Wisconsin (Carr & Stermer, 1952:48).
Because of poor economic conditions in the South, the migration north
continued after the war. Willow Run continued to operate under new
management, although for automobile manufacture rather than B-24
bombers. Because most people settled near the plant, located on the
east side of Ypsilanti, and because all Ypsilanti inhabitants were
aware of the presence of southern migrants, that part of the community
has become known as “Ypsitucky.” The salience of
Appalachians, demonstrated by the label “Ypsitucky,” makes
“Appalachian” available to these community members as an
identity marker 40 years after the mass migration of southerners.
Treatment of social network and linguistic variation has not been
carried out in settings where dramatically different dialects are in
contact. Ypsilanti is especially interesting, because the details,
including historical ones, of the two dialects and their own ongoing
internal changes are so well-known, as shown in the characterizations
of the Northern Cities Chain Shift and the Southern Chain Shift. In
fact, the results of the present study suggest that the effect of
social network may be more powerful than status in predicting dialect
acquisition in this community. This study is one answer to de Bot and
Stoessel's (2002:3) complaint that
“there is very little, if any, quantitative support for a direct
relation between social-network characteristics and language
use.”
This study aims to explore the speech of members of the migrant group
to ascertain whether they have acquired a particular feature of the
local dialect and whether their linguistic behavior is influenced by
their network affiliations. The hypothesis is the following: The
migrant inhabitants and their children whose social network ties are
dense and multiplex (Milroy, 1980) will not
have accommodated or will have accommodated least to the local norm
(the NCS).
METHODS
Respondents
Twenty-eight respondents (17 females and 11 males, ranging in age
from 28 to 81) were recruited at the Ypsilanti Township Adult Education
Center, Ypsilanti City and Township community centers and churches, and
through friends of friends. The Appendix shows each respondent's
pseudonym, sex, age, age at migration to Michigan, which state
she/he migrated from, her/his relationship to other
respondents, socioeconomic status, and Appalachian integration score
(explained below). Eight respondents were born in Michigan. Some of the
respondents belong to the same family and some are related by marriage.
The respondents in this study who were not born in Ypsilanti came from
southern states (mostly Kentucky) between the 1940's and
1960's. All of these respondents indicated that they came to be in
Ypsilanti for employment reasons (for themselves, their parents, or
their spouse).
Analysis of recordings
Vowel formant frequencies of word list data were obtained using a Kay
Elemetrics Computer Speech Lab model 4300B. Measurements were taken
from the steady state of the onset of each vowel.
1 Only the raising and fronting of the onset is treated
here.
Formant frequency data was normalized using
Plotnik.
2 Plotnik is a program developed by
William Labov at the University of Pennsylvania, which includes the
normalization system developed by Neary (1977).
To detect any divergence from a
nonshifted vowel, an index system was used that allows for the quantification
of the position of each vowel relative to other vowels in the
individual's system. Such index scores can show discrete
differences, visible only within individual systems, which are lost
when only vowel formant data are analyzed (
Evans
& Preston, 2000). Index score assignment involves
identifying a stable vowel, and comparing it to a vowel whose position
may have changed. Because
is said to change only in later stages of the NCS, as shown in Figure 2, it was chosen as a stable vowel for comparison
with
.
T-tests were performed on the normalized
mean scores of these and other such pairs of vowels. According to the
t-test results, a vowel was given a rating indicating its
position relative to a stable vowel in the system. For example, in the
case of
,
if the mean F1 of
is significantly greater than
,
then it was given an index score of one, indicating that it is
“lower” than
or “unraised.” If the mean F1 score of
is not significantly different from
,
it was given an index score of two and so on (see Table 1).
Index scores assigned for rating F1 means of
Table 2 demonstrates how index scores can
reveal the relative position of a vowel, whereas the actual formant
frequency means (even normalized ones) may be misleading. For example,
Laura and Darcy have greater F1 scores for
than Barbara and Anna. This leads to the erroneous conclusion that Laura and
Darcy have a lower
than Barbara and Anna, but t-tests on each respondent's
mean scores reveal that Barbara and Anna have a lower
(see Table 2) than Laura and Darcy, who have a
slightly raised
.
Index scores, therefore, are used for the statistical analysis of variables
and subsequent interpretation and discussion of the results. The index
scores and normalized mean scores were then analyzed using appropriate
statistical procedures to determine any significance of the independent
variables.
Comparison of index and F1 normalized mean scores of
for four respondents
Appalachian integration score
Following Milroy (1980:141), a numeric
calculation of each respondent's network was made. Throughout the
interview, the respondents gave information pertaining to their
friends, family, and coworkers. This information was used to calculate
the density and plexity of the respondent's network. Respondents
received points for having substantial kinship ties and coworkers in
the community. Information about the respondent's social network
was also collected overtly in the interview. Each was asked to indicate
approximately what percentage of their friends and associates were from
the South. For example, respondents reporting that 100% of their
friends were Appalachian received five points, and those reporting that
0% of their friends were Appalachian received zero points.
Thus, the Appalachian integration score (0 = lowest, 19 = highest),
parallel to Milroy's social network score, was obtained for each
respondent by combining the points for percentage of Appalachian
friends (0–5) and the points for density and plexity (0–5)
of network. The resulting Appalachian integration score for each
respondent is shown in the table in the Appendix.
RESULTS
Because fronting of
(a greater F2 mean score than found in older systems) is a characteristic of
the Southern Shift, this feature in these respondents' data cannot be
attributed to the NCS with any confidence. Therefore, only F1 of
,
a feature of the NCS but not the Southern Shift, is examined. The results of
the index score assignment for F1 of
,
as described previously, are shown in Table 3.
Nineteen of the 28 received an index score of 1 for F1 of
;
that is, these respondents did not demonstrate any evidence of a raised
as is found in the NCS. Nine of the 28 received a 2 for F1 of
,
that is, they have an F1 for
that is not significantly different from /E/ and is attributed to
influence of the NCS.
Index scores assigned to Ypsilanti respondents
Figure 4 provides an example of a
respondent with index scores of 1 for F1 and F2 of
.
George is a 46-year-old male who was born in Ypsilanti. His mean score
for F1 of
was significantly lower (as determined by a t-test) than that for
.
In addition, the
positions of his mid vowels,
,
were reversed as in the Southern Shift (Figure 3).
In short, George does not possess features of the NCS.
Results of acoustic analysis of data from respondent George.
Brenda, on the other hand, had an index score of 2 for F1 of
.
Her mean score for F1 of
was not significantly different (as determined by t-tests) from
,
as can be seen in Figure 5. Moreover, a comparison
with Figure 1 shows that even the position of
/a/ in Brenda's vowel system was more front (i.e., has a greater
F2) than
,
a feature of the NCS (see Figure 2). Brenda, also
born in Michigan, did not exhibit features of the Southern Shift, but did show
influences of the NCS.
Results of acoustic analysis of data from respondent Brenda.
A statistical analysis of these index scores was necessary to capture
any significant patterns among the vowel position scores and other
variables. An analysis of covariance (ANCOVA) was conducted to test the
relationship of
F1 index scores with sex, socioeconomic status, age, and Appalachian
integration, and to test for any interactions that may exist among the
independent variables (IVs). An ANCOVA was chosen because it allows for both
continuous and categorical variables in the model. The first step in the
analysis was to test if any interactions among the IVs existed. All
interactions were not significant at p < .05, therefore, the
interactions were left out of the final model. The final model tested
for the main effects of socioeconomic status, sex, age, and Appalachian
integration. The model itself was significant (df = 4,
F = 5.32, and p < .01). The IVs sex (p
< .05) and Appalachian integration (p < .05) were the
only significant contributors to the model. Each IV was also tested
separately against
F1 index scores with the appropriate statistical analysis. These results are
what follows.
Sex
Women have been found to be “leaders” of linguistic
change in many variation studies (Labov,
1994). This pattern has also been found in NCS. For example,
Eckert (1989) found that female respondents
in suburban Detroit had more raised tokens of
than the males. Ito (1999) also found that female
rural Michigan respondents lead in the use of raised
tokens. Therefore, it was expected that the women in this group would
have higher F1 and F2 index scores than the men. The result of a
chi-square analysis of sex and
F1 index scores showed that the difference between men and women was
significant (χ2 = 4.41, df = 1, p < .05)
with regard to
raising. A higher percentage of women (47% of women vs. 10% of men) received
an index score of 2 for
F1; that is,
of 47% of the female respondents was not significantly different from
in height. As expected, it appears that more women than men are raising
in this group.
Socioeconomic status
A chi-square analysis of
and socioeconomic status showed that the difference between the working-class
and middle-class respondents was not significant (χ2 = 1.47,
df = 1, p > .05). The division of these respondents into
only two social classes, middle and working, may have concealed discrete
differences among them (Labov, 2001:31),
however, the social structure of this group, originally working-class,
did not allow for the addition of a large enough set of
upper-middle-class respondents for comparison.
Appalachian integration
A Pearson correlation analysis was conducted to determine whether the
Appalachian integration score was significantly correlated with
raising. A significant negative correlation for social network and F1 score
was found (−.617, p = .001). That is, the higher the
Appalachian integration score (indicating tighter networks), the lower the F1
index score. Figure 6 shows the trend indicated by
the negative correlation; the respondents who received an index score of two
(i.e., those who raise
)
had lower Appalachian integration scores than those who received an index
score of one. Those respondents who had the highest Appalachian integration
scores, that is, those respondents with the tightest Appalachian social
networks, received an index score of one. It follows, therefore, that a tight
Appalachian social network in Ypsilanti serves as an inhibitor to adopting
features of the NCS. This result is consistent with the hypothesis that
members of tight Appalachian networks will not have acquired
raising. As described earlier, Milroy (1980)
showed that membership in a tight social network serves as a norm
enforcement mechanism and discourages adoption of features not
associated with the group.
raising according to Appalachian integration score.
Age
With regard to the age of a respondent and
F1 index scores, a Pearson correlation analysis was inconclusive, but Figure 7 shows that the majority of the older
respondents received an F1 index score of 1. Nevertheless, those
respondents who received an index score of 2 range from age 28 (the
youngest respondents in the study) to age 68. One might assume that age
of acquisition might be a more powerful predictor than age at the time
of investigation. Table 4 shows the
respondents' age in years at the interview and age at migration.
If it is assumed that being born in or early arrival in Michigan gave a
respondent an advantage for learning local norms, comparison of the
respondents who received an index score of 1 with those who received an
index score of 2 indicated that this is not always the case. Three
respondents who were born in Michigan and even some who arrived in
Michigan before age 16 received an index score of 1. It seems that age
is not as reliable a predictor for acquisition or lack of acquisition
of NCS features among Ypsilanti Appalachians as network, or sex.
Age and age of migration of respondents
raising according to age.
Social network, however, did correlate with age, suggesting, perhaps,
that a wider sample of more diverse age and network patterns might
reveal independent significance for both. When social networks and age
were analyzed, with a Pearson correlation analysis, they were found to
be positively correlated (r = .4, p < .03). That
is, the higher (older) the age of the respondent, the higher (tighter)
the Appalachian integration score. Older respondents had tighter
Appalachian social networks and perhaps older respondents were less
likely to demonstrate NCS features simply because of their age rather
than their tighter Appalachian social networks, but a larger sample
would be necessary to confirm this.
CONCLUSION
In summary, only 9 of the 28 Ypsilanti respondents possess
raising. Results of the statistical analysis of Ypsilanti respondents'
index scores for
show that sex and Appalachian integration are the most important factors
with regard to
raising.
Age, although it positively correlated with the Appalachian index
score, was also not significantly correlated with raising, but more
data from a wider variety of respondents would be required to reach a
conclusion about the differences resulting from age on the basis of age
of acquisition, as well as age at the time of investigation, and its
possible interaction with the Appalachian integration score.
These Appalachian migrants certainly would have benefited from
assimilating linguistically and socially to the local norms.
Documentation about Willow Run exemplifies the prejudice against
southerners that existed. Carr and Stermer (1952) described the events surrounding the
establishment of the Willow Run bomber plant: “Later because of
recruiting by the [Ford] plant, hillbillies from Kentucky,
Tennessee took over second place [in number of migrants] and
incidentally became ‘problems’ to the natives”
(1952:41). Even the authors, who were sociologists,
describe the migration as an “invasion” of
“hillbillies” (1952:41). Evidence from
folk linguistics indicates that the status of Appalachian speech and culture
does not fare much better today (Preston, 1996).
However, as the results above indicate, over half of these respondents, in
spite of the social repercussions, do not possess an important feature of the
local speech norm. Therefore, it is striking that Appalachian integration had
more influence on maintenance of Appalachian speech features than
socioeconomic status. Socioeconomic status has been a meaningful
variable in sociolinguistic research and has been found to be
significant in a multitude of studies. It is surprising, therefore, to
find that Appalachian integration is not simply an addendum to, but has
more explanatory value, than socioeconomic status in this study.
As Milroy (1980) showed in Belfast, these
results show the importance and relevance of social network relations
in explaining the different patterns of use of the local norm in this
group. Appalachian integration helps us understand the different
patterns demonstrated by Ypsilanti respondents who lived in Ypsilanti
from a very young age who did not demonstrate
raising, such as George (see Figure 4), Ray,
and Joe (see Table 4), and those who did
demonstrate
raising, such as Shelly, Mark, and Brenda (see Table
4). These results call for more exploration of the complexities and
influence of social network and socioeconomic status.
