Journal of Sociolinguistics 9/2, 2005: 225^253

Attribute networking: A technique for

modeling social perceptions1

Robin Dodsworth
The Ohio State University

An original, network-based technique is presented for modeling community

members’conceptions of local social space. Social categories derived from the
model are used to investigate the social meaning of linguistic variation. The
technique is first explained and then demonstrated using linguistic and eth-
nographic data from Worthington, Ohio, a Columbus suburb. Two linguistic
variables are analyzed: (1) /l/ vocalization; and (2) the phonetic realization of
the before vowel-initial words. The results are discussed in the context of
Columbus-area urban sprawl and its perceived threat to Worthington’s
distinctiveness.
KEYWORDS: Networks, social perceptions, social categories, /l/
vocalization, Columbus, Ohio

INTRODUCTION
Among sociolinguists, there is little if any doubt that some aspects of the way
people think about their local social surroundings interact with their linguis-
tic choices. The earliest example within the quantitative variation paradigm is
Labov’s (1963) analysis of (aw) and (ay) in Martha’s Vineyard, in which the
only significant social factor is orientation to island culture. While Labov
developed an understanding of this local dimension of identity by talking with
island residents, he asserts in a footnote that ‘discussions with community
leaders who were in a position to view [social patterns on Martha’s Vineyard]
as a whole’ proved more informative than conversations with the 69 speakers
(1963: 27).Yet the community leaders were themselves community members,
and their understandings of island social patterns undoubtedly stemmed from
lived experiences just as other informants’did. More recent studies ^ in parti-
cular, those that appeal to the community-of-practice concept ^ have gone a
step further by assuming from the outset that community members’ subjective
evaluations of local social space can help account for linguistic variation.
Community-of-practice studies interpret linguistic variation primarily in
terms of subjective experiences and individuals’ negotiations of personal and
group identities, as the boundaries of a community of practice are determined

# Published by Blackwell Publishing Ltd. 2005

9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden MA 02148, USA.
226 DODSWORTH

according to ‘criteria that are subjectively salient to the members themselves’

(Meyerhoff 2002: 533). Further, the community of practice is a tool for under-
standing the connections between individuals’concrete practices and broader
social patterns, a goal that is also apparent in Labov’s Martha’s Vineyard study.
Along those lines, Meyerhoff (2002: 543) describes the community of practice
as ‘an attempt to inform the general through the study of the particular.’
Specific social practices, and the local meanings they index, are viewed as
both instantiations of, and opportunities to understand, broader social phe-
nomena such as class. Eckert (2000: 40^41) observes that the community of
practice focuses ‘on the day-to-day social membership and mobility of the indi-
vidual, and on the co-construction of individual and community identity. In
this way, it ties social meaning to the grounded social aggregate at the same
time that it ties the grounded aggregate to abstract social structures.’ This
approach thus continues Labov’s early emphasis on local, subjective meaning
while aiming to make more explicit the connections between individual
(linguistic and non-linguistic) practice and larger social patterns.
Despite their explanatory potential at the level of the individual, current
ethnography-based studies of variation, rooted in subjective experiences, cannot
claim the level of replicability that quantitative studies generally can, and they do
not facilitate cross-community comparison. The type of information that many
qualitative studies take as basic not only must largely be inferred rather than
directly observed, but also tends (by necessity) to be culled and interpreted in a
relatively opportunistic, unstructured way. By contrast, the use of clearly
defined, observable social indicators such as income or education ^ in short, a
positivist approach ^ comes closer to being fully repeatable and allows linguistic
data to be interpreted in structured, pre-determined ways. In the social networks
framework (Milroy 1980, 1987; Milroy and Milroy 1992; Lane 1998), for example,
an individual’s position in a network is typically determined on the basis
of observed or reported ties with other members of the network. While
the community-of-practice framework is closely related, membership in a
community of practice is determined partly by group members’ fluctuating
conceptions of group boundaries.
Clearly, neither approach is sufficient on its own. An ideal approach to socio-
linguistic variation would combine the interpretive power of subjective percep-
tions with the replicability of quantitative data. This type of approach would
involve systematically uncovering and representing community members’sub-
jective conceptions of local social structures and processes. Subjective infor-
mation would thus be objectified in a sense, making it amenable to
quantitative analysis like social networks are. Such an approach would thus
maintain the importance of individuals’ everyday experiences while opening
the door to empirical justifications for delineating social categories and
abstracting from them.
Toward this end, this paper introduces an original network-based technique,
attribute networking, intended to systematically model community members’
# Blackwell Publishing Ltd. 2005
ATTRIBUTE NETWORKING 227

conceptions of local social phenomena, including any social boundaries or

events that are salient to them. In essence, perceptions of local social space are
operationalized in order to facilitate quantitative analysis of sociolinguistic
variation. The technique also has the potential ^ shared with the community
of practice ^ to shed light on perceived connections between local, concrete
social facts and broader social constructs (cf. Meyerhoff 2002; Milroy and
Milroy 1992). Attribute networking is not intended to, and cannot, replace cur-
rent approaches. Further, by introducing this technique, I do not intend to
imply that individuals enjoy perfect autonomy in making sociolinguistic
choices, that is that objectively ‘real’ social structures and processes (such as
wealth differentiation and globalization) have no influence (cf. Carter and
Sealey 2000 and responses in Potter 2000 and Fairclough 2000).
Attribute networking is first described and then illustrated with data from a
sociolinguistic study of Worthington, Ohio, a suburb of Columbus. Finally,
some of this technique’s potential strengths are discussed in the context of cur-
rent approaches to sociolinguistic variation.

ATTRIBUTE NETWORKING
The use of networks presented here is based heavily on Bearman, Faris and
Moody’s (1999) network-based representation of a series of events in the
Chinese Revolution according to segments of 14 life-story narratives. In their
model, nodes represent single events, and ties between nodes represent tem-
poral relationships. Similarly, attribute networks encode community mem-
bers’ stated conceptions of local social processes and categories and are
therefore not social networks: the nodes represent socially meaningful charac-
teristics of people in the community, and a tie between two nodes indicates the
perceived co-occurrence of the two characteristics that those nodes represent.
The social perceptions represented in the network are extracted from ethno-
graphic interviews during which informants are asked to talk about the com-
munity’s social space as they perceive it. For instance, the dyad in Figure 1
represents speaker A’s perception that people in her neighborhood (as opposed
to other neighborhoods in the community) tend to hold professional degrees.
In the context of attribute networking, the dyad in Figure 1means that one per-
son perceives a correlation between living in her neighborhood and holding a

live in A’s neighborhood

professionals

Figure 1: Two linked attributes

# Blackwell Publishing Ltd. 2005
228 DODSWORTH

professional degree. Crucially, it does not represent the perception that every-
one in the neighborhood holds a professional degree, nor that everyone in the
community who holds a professional degree lives in that neighborhood. In this
respect, attribute networks are similar to social networks: a tie between two
people in a social network does not indicate that they socialize exclusively
with each other, nor that they spend all or even most of their time together. In
social networks, the difference between strong and weak ties partially captures
the fact that ties are not identical in strength, but it remains understood that
two strong links need not be quantitatively or qualitatively identical.
Each community member’s perceptions are encoded in a single network. The
network need not be connected; that is, some nodes may not be reachable from
others. For instance, suppose that speaker A from Figure 1 states that most con-
servatives in the community (but not necessarily in her neighborhood) have
flags waving from their porches. Based on that information alone, the nodes
representing conservativism and flag-waving would not be connected to the
dyad in Figure 1.
The individual networks are intercalated to form an aggregate network. The
goal is to unite individuals’conceptions of the social space to form a system in
which diverse views are represented, points of agreement are apparent, and
the elements of the social space can be seen in relation to one another.
Intercalation consists primarily of two processes. First, if two networks have
the same set of nodes but do not share any dyads (i.e. no ties connect the same
two nodes in both networks), then the ties from one network can simply be
added to the other network. Second, if two networks share a dyad, then that
dyad’s link has a value of 2 in the aggregate network, reflecting the number of
speakers who expressed a perceived association between the characteristics
represented by the nodes. When two networks do not share any nodes, then
intercalating them is a matter of calling them a single, unconnected network.
The aggregate network is considered a rough model of the community’s
perceived social structure according to the informants whose interviews
contribute to it. Several characteristics of the aggregate network make it an
advantageous model. First, it can represent disagreement among community
members. If, contrary to the perception represented in Figure 1, a speaker per-
ceives that people living in A’s neighborhood tend not to hold professional
degrees, these conflicting perceptions are united by adding a node labeled
something to the effect of ‘no professional degree’ and linking it to the node
labeled ‘live in A’s neighborhood.’ To the extent that community members have
conflicting conceptions of local social phenomena, this feature of attribute
networking is necessary.
Another potential advantage is that the model does not require discrete
social categories a priori because all nodes may be connected. In the
Worthington aggregate network, the largest component (set of connected
nodes) includes nodes that represent relative affluence and nodes that
represent relative poverty, as well as nodes that represent a diverse range of
# Blackwell Publishing Ltd. 2005
ATTRIBUTE NETWORKING 229

locations and levels of engagement in the community. The fact that all of these
nodes are reachable from one another suggests that they are separate yet con-
nected parts of the same fluid social system, just as members of a social network
may be more or less central but still linked, directly or indirectly. Although this
characteristic is not fully exploited in the illustration below, it has the potential to
facilitate a treatment of social categories as fluid, overlapping entities with
dynamic, negotiable boundaries, as in the community-of-practice tradition.
A critical assumption underlying attribute networking is that nodes or subsets
of nodes which are structurally important in the aggregate network are the
most salient aspects of social identity in the community. Several quantitative
measures of node importance, all of them focusing on individual nodes rather
than sets, are illustrated in the following section. Structurally important nodes
are likely to represent characteristics that many community members consider
socially meaningful, because they are mentioned by multiple speakers and/or
are linked to multiple nodes. Other nodes are likely to be connected to one
another only through these important nodes, except in particularly dense net-
works. For those reasons, the characteristics associated with important nodes
are good candidates for social variables in quantitative analysis.

AN ILLUSTRATION OFATTRIBUTE NETWORKING

In this section, attribute networking is illustrated with a sociolinguistic study
of Worthington, Ohio, a small town north of Columbus whose distinctiveness
is increasingly threatened by Columbus sprawl. As this type of social context
is quite familiar, and as the linguistic results present an expected picture of
resistance to urban sprawl, the Worthington study is not itself offered as a
contribution to variation research but serves only to clarify the network
approach described above.

Worthington
In 1803, settlers from Massachusetts and Connecticut founded the village
of Worthington in central Ohio beside what is now called the Olentangy
River, modeling it after a traditional New England village (McCormick and
McCormick 1992). Nine years later, the city of Columbus was established
south of Worthington. A locus of vigorous urban expansion, Columbus has
crept northward to the point that it now completely surrounds Worthington.
During the 1970s, Columbus and Worthington agreed to define the latter’s
school district boundary such that it encompassed not only the city of
Worthington but also some of the surrounding areas belonging to Columbus.
Since then, countless residential neighborhoods have been developed in
those areas, where the relatively high Worthington taxes and property values
do not apply despite their inclusion in the prestigious school district.

# Blackwell Publishing Ltd. 2005

230 DODSWORTH

According to the Worthington school district’s website, the district’s popula-

tion grew from 28,648 in 1980 to approximately 58,000 in 2000. Unlike
20 years earlier, the majority of students attending the well-funded
Worthington City Schools live in Columbus, and two-thirds of the district’s
school buildings lie on Columbus land.Yet Worthington, now a city of close to
15,000 people compared to over 700,000 in Columbus, remains somewhat
politically and socially distinct despite being swallowed by urban growth.
From an outsider’s perspective, Worthington is the prototype for an
upper-middle-class suburb. Ninety-four percent of its residents identify as
white, in contrast to 67.9 percent in Columbus, and its 2000 median annual
household income of close to $70,000 dwarfed the Columbus median of
under $40,000 (U.S. Census Bureau 2000). Perhaps the most important
component of Worthington’s distinctiveness, however, is what many
residents call its ‘colonial feel,’ or the collective consciousness of its New
England roots. Old Worthington, the space occupied by the original 1803
village, is governed by a set of strictly enforced architectural guidelines
geared toward maintaining a colonial atmosphere. A four-quadrant village
green lies at the center of OldWorthington, and several of the New England-style
buildings constructed by the founding settlers still stand, some of them
maintained by the Worthington Historical Society. Regular public events
refer to the city’s heritage: the annual Founder’s Day, for example, celebrates
the establishment of the village and involves a week-long educational
program for elementary schools.
Yet these traditions preserve not historical knowledge but collective
imagination. Helen, a Worthington resident in her 80s who has served as
the historical society’s curator, described Worthington’s historical
awakening:

1. Helen: It’s become very conscious that it was a transplanted New England
village.
RD: It’s become conscious?
Helen: Yes.
RD: You mean it wasn’t conscious before?
Helen: Uh, it wasn’t conscious until 1926 when the country had its
hundred and fiftieth anniversary, and all of a suddenWorthington said
‘Hm, we have a village green, and we have some nice old buildings.’
And the, it was beginning then until the early thirties that the
Kilbourne school and the Presbyterian church and the library were
built, and they were built after earlyAmerican structures . . . and at that
point, I think it was in the early 1930s that they started a historical
society that didn’t get anywhere but then in um, by the 50s they were
really very much aware and with, thenWorthington had its hundred
and fiftieth anniversary, and they did a pageant, and people came,
became very much aware of buildings that should be saved.
...

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 231

The city itself is very conscious of what being a pioneer village, what
the appeal is to people, and when they decided that it was good
monetarily, then they all pitched in. (laughter) But Worthington has a
draw because of that.
The city’s historic consciousness has more to do with the benefits of adopting a
distinctive identity than with any continuous inheritance of tradition from
1803. Most of the current residents of Worthington, however, were born after
the 1930s when the interest in colonial New England culture took hold. For
them, Worthington’s historic identity is genuine and is an unquestioned com-
ponent of the local culture. Even so, it is part of Worthington’s local prestige
and small-town appeal, and high property values are one by-product. It is no
secret that the material elements of Worthington’s culture are purposefully
and consciously maintained. A column on the opinion page of an April, 2004
issue of one of Worthington’s two weekly newspapers offers a pointed
illustration:
2. Take a drive down High Street or through the streets of Old
Worthington, compare those streets to any nearby community, and you
cannot help but be impressed.
That taste and character reign is no accident. The Worthington
‘look’ has actually been carefully orchestrated by a board of dedicated
public servants. (Brooks 2004: A4)
Worthington residents necessarily share public spaces with people living in
Columbus and nearby suburbs, especially in the school system, but for many of
the community members, the city boundary marks a real and salient division.

The data
The linguistic and social data presented here are extracted from a total of 24
ethnographic interviews, each of which lasted approximately one hour. The
speakers lived either in Worthington or in surrounding areas included in the
Worthington school district. At the time they were interviewed, one speaker
was 15 years old, five were in their early 20s, five were at or near 40, nine
ranged from mid-50s to early 60s, and three were 80 or older. All speakers
younger than 30 grew up inWorthington and all speakers over 30 had lived in
Worthington for at least 20 years. A ‘friends of friends’ strategy (Milroy 1980,
1987) was employed to select speakers, beginning with two speakers who did
not know one another and lived in different parts of the community. Speakers
were asked what types of people, social groups or divisions there were in
Worthington, or what types of people. This open-ended question is similar to
one that Mendoza-Denton (1997: 71) asked Latina girls in a Silicon Valley high
school: ‘What other groups of Latina students do you think there are here at
Sor Juana High School, and who do you know that might be in these groups?’
The question took a slightly different form with each informant, but was

# Blackwell Publishing Ltd. 2005

232 DODSWORTH

generally clarified with the example of a stereotypical high school and its
social groups. The representation of speakers’ stated perceptions of
Worthington’s social space is illustrated here with an excerpt from an inter-
view with Ann, who is in her mid-50s and lives near Old Worthington (the
space occupied by the original village):
3. There are a lot of people that are educators that live [in Worthington
proper], a lot of professors at OSU . . . it’s mostly professional people.
Now, that said, in the outskirts of Worthington which is considered
Columbus land but Worthington schools, there are a lot, and where
most of the apartments are, I would say that that’s where a lot of the
single-parent families live, and they would not necessarily be the
professionals but more of the clerical or even factory workers . . . That
would be the non-professionals, a lot of, in those areas . . .
The bolded words in Example 3 map onto attributes in Table 1. In the table, a 1
represents a stated association between two attributes, and a 0 represents the lack
of a stated association. In the individual attribute networks, ties are valued either
1 or 0 (and cells in the diagonal of matrices have the value 0). In the aggregate
network, however, a tie’s numerical value reflects the number of individual
networks in which it appears, which is equal to the number of speakers who stated
the connection. Thus the maximum possible value for any tie in the aggregate
network is simply the total number of informants, or individual networks.
Some potential attributes from Example 3, such as being a clerical worker,
are not represented in the matrix; the level of detail used in representing the
spoken discourse as a matrix can obviously vary. In Ann’s case, it seemed that
clerical and factory workers were merely examples of the category ‘non-
professional.’

Potential weaknesses of the technique

Clearly this way of eliciting and encoding subjective information, and the
assumption that a set of connected nodes represents a speaker’s conception of

Table 1: Matrix of attributes from Example 3

W Non- Single
proper Professionals Outskirts professionals parent Apartments

W proper 0 1 0 0 0 0
Professionals 1 0 0 0 0 0
Outskirts 0 0 0 1 1 1
Non- 0 0 1 0 1 1
professionals
Single parent 0 0 1 1 0 1
Apartments 0 0 1 1 1 0

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 233

the social structure, come with a set of complications. Perhaps the most
obvious and stubborn problem is that speakers are unlikely to talk about their
perceptions of the social space in a way that translates easily into sets of nodes
and ties. The excerpt from Ann’s interview is unusual in the relatively clear
way it categorizes people according to concrete attributes such as job type and
location of residence. Many other speakers described sets of social attributes
but either did not link them to one another very explicitly or did so with consid-
erable hedging and uncertainty. Further, many speakers discussed sets of
social characteristics that were not easily represented with discrete nodes. For
instance, one speaker described the economic status of three Worthington
neighborhoods relative to one another, suggesting a scale of wealth: according
to him, ‘Old Worthington’ was the least affluent neighborhood, followed by
‘KilbourneVillage,’ followed by ‘Worthington Hills,’ the most affluent. In princi-
ple, this idea can be captured in a network by arranging nodes hierarchically,
but the challenge is to do so in the context of the larger system of nodes and
ties in which the hierarchy is embedded. I handled this particular case rather
awkwardly, by representing each neighborhood as a node connected to
another node that described its economic status relative to the other neighbor-
hoods (highest, middle, or lowest). It is also important to note that not all rele-
vant information from ethnographic interviews is easily represented in
network form. For instance, some Worthington informants described what
they perceived as others’conceptions of the city’s social space, but emphasized
that they did not share these perceptions. This kind of disclaimer is important
but difficult to encode in a network. Further, representing only perceptions
that have been explicitly stated, as has been done here, entails that all implied
information is excluded from the network. For instance, speakers are likely to
imply a constrastive pair with the mention of only half of the pair, as in the
excerpt above where Ann associates single parents with Worthington’s out-
skirts but does not explicitly link the two-parent families with Worthington
proper. These weaknesses underscore the importance of coupling attribute
networking with ethnographic observation. Other complications, including
the probability that no speaker is likely to mention all the attributes that he/
she considers socially salient, are minimized by combining the individual net-
works into an aggregate network.

Structural characteristics of the aggregate network

Three of the interviews that contribute linguistic data to this study were con-
ducted after the network aspect had been completed, so attribute networks
were not constructed for them. Intercalating the individual networks from the
remaining 21 interviews yielded an aggregate network with 138 nodes, pic-
tured in Figure 2 with selected nodes labeled. The network consists of one
large component containing 100 nodes, and 14 smaller components. Among
the smaller components, there are no ties valued greater than 1, a sign that
# Blackwell Publishing Ltd. 2005
 234

live in W proper

sense of community

Colonial Hills

live in Old W

no community involvement

‘grand’ houses
live in apartment
outskirts (N Columbus)
single parent
DODSWORTH

# Blackwell Publishing Ltd. 2005

Figure 2: The Worthington aggregate network (selected nodes labeled)
ATTRIBUTE NETWORKING 235

they represent connections that are not broadly recognized in the community.
The large component can be viewed as having two sections which are con-
nected only through the nodes labeled ‘live in Worthington proper’ (the space
inside Worthington’s political boundary) and ‘sense of community.’ The larger
of the two sections roughly corresponds to living in what some informants
referred to as ‘the outskirts,’ or the areas of Columbus lying inside the
Worthington school district. The smaller section roughly corresponds to
living in Worthington proper.
A striking characteristic of the network is its low number of ties valued
greater than 1 ^ recall that a tie’s value indicates the number of speakers who
have stated an association between the nodes it connects. Only 17 ties have
values greater than 1, and the highest value is 5 out of a possible 21 (the num-
ber of speakers). This characteristic has multiple interpretations. It may, for
instance, indicate disagreement: if all informants had been in perfect
agreement as to the community’s social structure, then many ties would have
values near the maximum of 21, the only low values resulting from differences
in the ways that informants stated their perceptions.2 For instance, the
node labeled ‘live west of the Olentangy River’ lies adjacent to nodes labeled
both ‘wealthy neighborhood’ and ‘live in an apartment,’ and the node
labeled ‘live in an apartment’ is adjacent to several nodes having to do with
low socio-economic status. This configuration reflects speakers’ differing
overall impressions of the outskirts: some speakers broadly conceive of the
area as a wealthy suburban space, while some view it as less affluent
than Worthington and beset by more crime. Many speakers perceived
socio-economic differences among specific places in the outskirts, which are
also represented in the aggregate network. One implication of this range of
views is that the areas in the outskirts are in social flux as suburban
development continues and Worthington residents have more and different
kinds of contact with Columbus residents. Thus the perceived social system is
constantly changing, with its various social ‘categories’ being defined and
redefined with respect to one another. As Mendoza-Denton (1997: 37) says,
‘the social categories here described are not conceived of as bounded sets, but
rather as parts of a single coherent system that is always fluid, always
changing, precariously equilibrated, and constantly innovating on itself.’
Another potential source of low-valued ties is disparity among informants
with respect to what they remembered to say about the community or were
willing to say.

Finding important nodes

The following measures of node importance were employed to find nodes in the
aggregate network that represented locally salient social characteristics. All
calculations were performed by the network software UCINET (Borgatti,
Everett and Freeman 2002).
# Blackwell Publishing Ltd. 2005
236 DODSWORTH

Betweenness centrality. A node’s betweenness centrality, a particularly

important measure of structural importance (Freeman 1977), is calculated on
the basis of the number of geodesics, or shortest paths between two nodes,
that it lies on (Wasserman and Faust 1994). Nodes with high betweenness cen-
trality are responsible for uniting other nodes. For example, because the node
representing living in Worthington’s outskirts lies on the geodesic between the
nodes representing more crime and paying Columbus taxes, the latter two
attributes can be assumed to be (subjectively) related to one another only insofar
as they are both properties of living in the outskirts. The two nodes with the
highest betweenness centrality, with roughly equal scores, corresponded to
living in Worthington proper and having a sense of community (see the labeled
nodes in Figure 2).

Degree. As in social networks, degree refers to the number of ties incident to a

node, or the number of nodes it is adjacent to; a node directly connected to five
other nodes has a degree of five. Nodes with high degrees are likely to represent
attributes that have been mentioned by more than one speaker, so degree is an
indicator of how broadly recognized a given attribute is. The node with the
highest degree corresponded to living in an apartment, followed by living in
Worthington proper and living in the outskirts.

Incidence with at least one tie valued greater than 1. In attribute networking, a
tie’s value reflects the number of individual networks it appears in, or the num-
ber of speakers who have expressed the association. The higher a tie’s value,
the more likely it reflects a perceived social fact that is recognized throughout
the community rather than just by a few speakers. Nodes incident to high-
valued ties therefore represent attributes that have been mentioned by multiple
speakers in connection with the same attributes. This indicates that the attri-
bute’s local meaning is similar for multiple speakers. The node incident to the
greatest number of valued ties, and also the greatest-valued tie, corresponded
to living in an apartment.

Cutpoints. Cutpoints are nodes whose deletion would create more compo-
nents (distinct sets of connected nodes) in the network. Therefore, a cutpoint
holds together at least two sets of nodes that would be disconnected from one
another without that node (Wasserman and Faust1994). In the context of attri-
bute networking, cutpoints are important because they may connect densely
linked clusters of nodes that represent separate ideas. An interesting cutpoint
in the aggregate network is the node representing having a sense of commu-
nity, which connects the sections roughly corresponding to living in
Worthington proper and living in the outskirts, respectively. Cutpoints are
never at the extreme periphery, though not every cutpoint is a particularly
central node.
# Blackwell Publishing Ltd. 2005
ATTRIBUTE NETWORKING 237

Discrete coreness. Discrete coreness treats a network as consisting of two

classes of nodes, core and peripheral. In an ideal core/periphery structure,
core nodes are maximally connected, while peripheral nodes are adjacent to
some core nodes but not to other peripheral nodes. Core nodes therefore hold
the network together and may determine its shape. Coreness is similar to cen-
trality insofar as all core nodes are relatively central; however, not all central
nodes belong to the category of core nodes, especially when the network has
multiple centralized sets of nodes, as the aggregate network does (Borgatti and
Everett 1999). Discrete coreness turned out not to be a particularly useful
metric for evaluating this aggregate network because only three nodes, form-
ing a triad, were classified as core: living in the outskirts, living in an apart-
ment, and single-parent families. This result obscures the fact that some nodes
in both sections are much more peripheral to the network than others.

Continuous coreness. Continuous models of coreness treat nodes as having

varying degrees of coreness based on the strength of their ties. A node whose
ties have high values thus has a greater degree of coreness than a node with
low-valued ties (Borgatti and Everett 1999). This metric produced a more inter-
esting result than the discrete coreness metric because all nodes received a
coreness score; a node was considered core if its score was at least one standard
deviation above the mean, for all nodes in the aggregate network. The 11 nodes
selected via this method are nearly a proper subset of the nodes incident to ties
valued greater than 1. Significantly, the node representing single parents was
not selected by this method, though it was one of only three core nodes accord-
ing to the discrete model of coreness.
Table 2 shows the number of nodes selected by each metric, and the criteria
for node selection in each case. The set of nodes selected by these measures
was narrowed to the set of nodes that were structurally important enough to
be selected by at least five of the six measures (to allow for the incompatibility
of discrete coreness with the aggregate network). The resulting set contains
seven nodes:
1. live inWorthington proper (within the city’s political boundary)
2. live in the outskirts (outside the city’s political boundary but inside the
Worthington school district)
3. live in Old Worthington (the space where the original village was
established in 1803)
4. little or no community involvement
5. live in Colonial Hills (a neighborhood inWorthington proper)
6. live in a‘grand’ house
7. live in an apartment.
These attributes were taken to be the most salient and broadly recognized
among the informants. For that reason, it was hypothesized that they would

# Blackwell Publishing Ltd. 2005

238 DODSWORTH

Table 2: Finding structurally important nodes

Measures of node Criteria for node # of selected
importance selection nodes

Normalized betweenness At least one standard deviation 11

centrality above the mean
Normalized degree At least one standard deviation 10
above the mean
Incident to at least one tie All nodes incident to a tie valued 21
valued greater than 1 greater than 1
Cutpoints [determined by model] 21
Coreness, according to discrete [determined by model] 3
core/periphery model
Coreness, according to continuous At least one standard deviation 11
core/periphery model above the mean

interact in some systematic way with sociolinguistic variables inWorthington.

To test this hypothesis, the first four attributes were used as the basis for con-
structing independent social variables for a VARBRUL statistical analysis.
(Two of the remaining three (live in Colonial Hills, live in an apartment)
describe none of the informants. The remaining attribute (live in a ‘grand’
house) was too vague and subjective to permit confident categorization of
speakers.) The first independent variable corresponded to location of residence
and had three variants: (1) Old Worthington (six speakers); (2) Worthington
proper but outside Old Worthington (11 speakers); (3) the outskirts (seven
speakers). The second independent variable encoded whether the speaker was
generally involved in the Worthington community, according to the speaker’s
own estimation (13 involved, 11 not involved). None of the speakers had signifi-
cant involvement in the Columbus community. Finally, independent variables
were established for sex (15 female, nine male) and age. These factor groups
were included in both VARBRUL analyses of linguistic variables discussed
below.

Linguistic variables
The linguistic data used in the following investigations of two linguistic vari-
ables was extracted from the same ethnographic interviews that yielded the
attribute networks.

Variable 1: /l/ vocalization. Post-vocalic (dark, coda) [
] differs from pre-
vocalic (light) [l] in that it involves greater retraction of the tongue body and
a delayed raising of the tongue tip relative to retraction of the tongue body

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 239

(Sproat and Fujimura1993). In several varieties of English, the tongue tip may
be raised only minimally, resulting in little or no alveolar contact and produ-
cing a ‘vocalized’ variant (Hardcastle and Barry 1985). Horvath and Horvath
(2002: 324) note that this vocalized /l/ is a back vowel or semivowel which
may be rounded and/or labialized; a variety of realizations, including [G] and
[U], has been observed in British dialects (e.g. Wells 1982). In their study of
speakers of various dialects of British English and one speaker of Australian
English, Hardcastle and Barry (1985) observe that the vocalized variant
occurred more often when followed by a velar consonant than by a palato-
alveolar consonant, and least often when followed by an alveolar consonant.
Further, preceding front vowels favored vocalization to a greater extent than
back vowels did. Similarly, Horvath and Horvath (2002) found the vocalized
variant most often in varieties of Australian English when the preceding
vowel was high, front, or long. Carver (1993) reports the absence of [l] and
weak [l] before final consonants in several Wisconsin counties, and Labov
et al. (1968) observe /l/ vocalization in NewYork City.
The vocalization of /l/ is a feature common to many central and southern
Ohio dialects. Ash (1982a, b) reports the vocalization of both syllable-final
and intervocalic /l/, resulting in a (possibly rounded) voiced glide, in
Midwestern cities including Columbus, and Durian (2004) observed /l/ voca-
lization among AAVE speakers in Columbus. A linguist from outside Ohio
remarked that a Columbus radio station advertises the ‘gowden owdies’ (i.e.
golden oldies). Ash (1982b: 10^11) claims that /l/ vocalization did not origi-
nate in Philadelphia until after the early 1940s, and that it is a new dialect
feature along the Eastern Seabord generally. Allen (1976; reported in Ash
1982b: 9) found infrequent pre-consonantal /l/ vocalization in Northern
and Midland speech. Given these facts, it is assumed here that in central
Ohio, the vocalized variant is the newer form.
The present study of coda /l/ vocalization in Worthington was performed
using 724 tokens extracted from 21 interviews. The set of tokens includes /l/
occurring syllable-finally as in all or almost, as the first segment in a coda con-
sonant cluster as in cold, and instances of syllabic /l/ as in little.With the inten-
tion of controlling for stylistic effects, only the first 20 minutes of each
interview was used, during which the speaker was either talking about how
long he/she had lived in Worthington or was still engaged in casual, introduc-
tory conversation.
Most studies of /l/ vocalization consider only two variants, vocalized and
unvocalized. Yet just as [l] and [
] are not categorically distinct entities (Sproat
and Fujimura 1993), the vocalized and unvocalized variants of post-vocalic /l/
clearly encompass overlapping sections of a continuum, as Hardcastle and
Barry’s EPG analysis demonstrates. It follows that categorizing tokens as voca-
lized or unvocalized is not entirely straightforward, unless objective acoustic cri-
teria are used (and even they will be somewhat arbitrary). Simply listening to
tokens carries the danger that the listener’s category boundary between [
] and
# Blackwell Publishing Ltd. 2005
240 DODSWORTH

the vocalized variant is either too narrow or too broad to capture important var-
iation patterns. Further, few listeners could classify tokens with perfect consis-
tency, even in the somewhat unlikely case that they find it easy to decide
whether each token is vocalized or unvocalized.
In an effort to sidestep these problems, three linguists other than myself
coded the data, all graduate students who are native English speakers, have
had training in phonetics, and do not vocalize /l/. The listeners independently
coded each token as [
] or an unspecified vocalized variant. In the VARBRUL
analysis, a token was coded as vocalized if at least two of the three listeners
judged it as such. The data are summarized in Table 3. In addition to the social
variables identified above, the tokens were coded for the following linguistic
factors:
1. morpheme-final vs. morpheme-internal (most instances of syllabic /l/
were morpheme-final)
2. preceding segment: labial, coronal, or dorsal consonant; front, back, or
central vowel
3. following environment: labial, coronal, or dorsal consonant; front, back,
or central vowel; pause.
Preceding segment, following segment, and location of residence emerged as
the significant factor groups. Results are given in Table 4, with percentages for
sex and age given at the bottom of the table. As shown, the closer a speaker
lives to Old Worthington (or ‘downtown Worthington,’ as some speakers
called it), the less likely he/she is to vocalize /l/. Cross-tabulating location of
residence with the phonetic factors, shown in Figures 3 and 4, provides a
more textured understanding of all three factor groups. Speakers from Old
Worthington and Worthington proper follow nearly identical patterns of
vocalization with respect to the preceding segment, with the exception of
central vowels. Columbus speakers, however, follow a strikingly different
pattern, in particular with respect to their high rate of vocalization after dorsal
consonants. It is interesting that Columbus speakers vocalize at close to the
same rate as the other speakers when the preceding segment is a coronal con-
sonant or a front vowel even though their overall pattern is quite different.
Turning to Figure 4, while back and central vowels encourage vocalization
among Columbus speakers, there is little difference among the other segments.

Table 3: Listeners’ judgments of /l/ tokens

Category Number % total

Vocalized according to at least two listeners 114 16

[
] according to at least two listeners 610 84
Vocalized according to all three listeners 36 5
[
] according to all three listeners 380 52

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 241

Table 4: Factors conditioning /l/ vocalization

(input ¼ .120, p < .005)
Factor (N) Factor weight % vocalization

Preceding segment
labial C (114) .650 21
coronal C (152) .376 9
dorsal C (26) .339 7
front V (199) .296 6
backV (219) .696 25
central V (14) .741 28
Following segment
labial C (123) .657 24
coronal C (333) .446 12
dorsal C (35) .781 31
front V (57) .396 10
backV (28) .583 17
central V (28) .162 3
pause (106) .525 16
Location of residence
Old Worthington (110) .293 7
Worthington proper (429) .506 15
outskirts (185) .615 21

Sex (not significant) % vocalization

female (463) 17
male (260) 12
Age (not significant)
15 (33) 18
20s (182) 15
30s (182) 15
late 40s^ early 50s (181) 16
60s (75) 14
late 70s^ early 80s (70) 12

By contrast, dorsal consonants favor vocalization for Worthington speakers,

while back and central vowels do not. With respect to both preceding and fol-
lowing segments, Old Worthington speakers and Worthington proper speakers
pattern together, showing marked differences from the Columbus pattern. I
return to this fact below.

Variable 2: (the). The second variable discussed here deals with the realiza-
tion of the vowel in the definite article the before vowel-initial words: [Dij{p /]%
versus [DV{p /]% for the apple.While only two variants, [DV] and [Dij], were distin-
guished in this analysis, the first variant is sometimes followed by a glottal
# Blackwell Publishing Ltd. 2005
242 DODSWORTH

45

40
Percentage /l/ vocalization

35

30

25 Columbus
W proper
20
Old W
15

10

0
labial C coronal C dorsal C front V back V central V
Preceding segment

Figure 3: Interaction between preceding segment and location of residence

stop. It is assumed that the older and prescriptively standard variant is [Dij],
though both variants occur in formal contexts. Unlike /l/ vocalization, this
variable has not been previously documented in central Ohio. The data come
from 21 interviews; all tokens from each interview were used, yielding a total
of 302. In addition to the social factor groups used in the /l/ vocalization analy-
sis, they were coded for the following linguistic factor groups:

50

45
Percentage /l/ vocalization

40

35

30
Columbus
25
W proper
20 Old W

15

10

0
labial C coronal C dorsal C front V back V central V
Following segment

Figure 4: Interaction between following segment and location of residence

# Blackwell Publishing Ltd. 2005
ATTRIBUTE NETWORKING 243

1. place of the following vowel

2. tenseness or laxness of the following vowel
3. stressed/unstressed following syllable (secondary stress was coded with
primary stress).
Only the first group, place of the following vowel, was significant in the first
VARBRUL run. Community involvement surfaced as the only significant social
group other than age, which was significant only when coded with two broad
categories: (1) early 40s and under; (2) mid-50s and over. (No informants fell
between these categories.) However, some of the cells corresponding to the
older speakers contained very little data, so the model is probably not reliable
with respect to age. Factor weights are given in Table 5. The relative factor
weights for place of the following vowel are somewhat difficult to interpret,
but it appears that front vowels generally favor [DV], possibly because it avoids
the juxtaposition of two front vowels. It is clear, however, that speakers with com-
munity involvement use the older variant more often. The cross-tabulation
of the two factors reveals that speakers with and without community involve-
ment show nearly identical patterns of variation with respect to the place of
the following vowel, as shown in Figure 5. While the place of the following
vowel conditioned the two groups of speakers’ variation in the same way,
speakers with community involvement used the older, more standard variant
significantly more often than those without community involvement.
Thus the pattern of variation with respect to linguistic context is uniform
throughout the community, but the level of variation is socially differentiated.
These results can be connected to the /l/ vocalization results by comparing
the set of people living in Old Worthington or Worthington proper with the
set of people who have community involvement, as shown in Table 6. Of the

Table 5: Factors conditioning (the): (app. value ¼ [DV], input ¼ .190,

p < .005)
Factor (N) Factor weight %[DL]

Place of following vowel

high front (32) .580 25
mid front (65) .702 45
low front (31) .638 39
mid central (68) .257 15
mid back (59) .432 22
low back (47) .524 23
Community involvement
community involvement (191) .348 16
no community involvement (111) .693 48

# Blackwell Publishing Ltd. 2005

244 DODSWORTH

80
70
Percentage [DV]

60
No community
50 involvement
40
Community
30 involvement
20
10
0
nt

ck
nt

ra
nt

ck
fro
fro

ba
nt
fro

ba
ce
id

id
gh

w
m

m
id
lo

lo
hi

Figure 5: Interaction between place of following vowel and community

involvement

13 speakers who have community involvement, only one lives outside

Worthington, so the two sets are nearly coextensive. Most of the speakers
who use [DV] the least are therefore also the speakers who vocalize /l/ the
least.

Discussion of results
The two linguistic variables show essentially the same pattern: speakers who
live in the city of Worthington, or who have community involvement, use the
older variants more frequently than speakers who live in Columbus or have lit-
tle or no community involvement. Although the attribute networking techni-
que correctly identified location of residence and community involvement as
parameters that interact with linguistic variation, it does not directly explain
the interaction. Even so, the aggregate network does speak to the connection
between location and community involvement because it shows direct
and indirect connections among attributes, thereby taking a step toward
explaining their contextual importance. In the Worthington aggregate net-
work, there is a node labeled ‘no community involvement’ ^ one of the seven

Table 6: Intersection between Worthington residences and community

involvement
Speakers living in Old Worthington or Worthington proper 14
Speakers who have community involvement 12
Speakers who live inWorthington and have community involvement 11

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 245

structurally most important nodes ^ as well as a node labeled ‘community

involvement.’ Both of these nodes are adjacent to nodes representing locations:
having no community involvement is adjacent to ‘outskirts,’ ‘west of the
Olentangy River,’ and ‘Worthington Hills’ (a neighborhood in the outskirts);
and ‘community involvement’ is adjacent to ‘Worthington proper.’ Thus the
relationship between community involvement and place of residence is well
represented in the network, reflecting the fact that many speakers addressed
the relationship between geography and community during their interviews.
One of the most elaborate discussions of this theme was provided by Rita, who
lives in Old Worthington and has the most community involvement of any of
the speakers:
4. Rita: Um, I guess I see the boundaries of Worthington as, um, the
communities that do get involved, the parts of the communities that do
get involved. So I don’t see West Worthington which is up by Sancus
and all the apartments up there, I don’t see um, uh, Worthington Hills,
uh, because uh, those are either not folks who who want to be in a
community or folks who want their own community.
...
And now I think you know there are no boundaries between
Worthington and Polaris.3 I mean it’s all, there’s no physical
definition. And the way things uh, in Columbus, and therefore in
Worthington, are growing and developing and expanding as, as far as
they can go I don’t think there is gonna be anything that that defines
one community to the next. Um, and I guess that’s just, that’s just city
development, I don’t know. I don’t know, pretty soon we’re gonna
look like Tokyo, where, one, one city just, just flows right into the next.
RD: How do you feel about that?
Rita: I’m not keen on it. We’re too big. We’re too big, already. Because
we don’t know each other. Because we’re not, involved, you know
because, because you can now say ‘Oh well this is theWorthington
school district,’ but you pay Columbus taxes, but are you, are you affiliated
withWorthington? Are you, you know, are you involved? And I think maybe
it’s because they’re too far out and, and don’t feel a connection, I don’t know.
Rita links urban expansion to the loss of community identity in Worthington
and in the Columbus area generally. For her, one of the indicators of waning
community identity is the absence of community involvement, and urban
growth is partly to blame for this absence because it erases boundaries
between communities and creates residential areas that are distant from the
centers of community activity. Other speakers draw similar connections. Greg,
for instance, who lives just outside Old Worthington and has strong civic invol-
vement, links the products of urban expansion and community involvement,
but with emphasis on the political boundary:
5. Greg: I do sense, or perceive that there are the, a lot of people enjoy
and want to live in this area because of the Worthington schools, and

# Blackwell Publishing Ltd. 2005

246 DODSWORTH

that’s an attraction. But there does seem to be a difference in people

who live within the city limits of Worthington versus those who live in
Columbus but are in the Worthington school district. I don’t sense,
and this is just, this is just the, a perception, it’s not, nothing to back it
up. It’s just the, they don’t have the same loyalty or closeness with the
same people who live in the city proper.
...
See, we lived [in the outskirts] for a while.We were in that 43235
West Worthington when we lived on Satterfield, and uh, I don’t think I
ever lost the sense that I was, uh not part of Worthington because I still
worked inWorthington, still did the uh announcing and everything for
the Worthington schools, still did Worthington Resource Center and
things like that, so I didn’t feel it so much. I guess I always felt like I
was gonna come back in the political, uh,Worthington anyway.
Like Rita, Greg links community involvement to membership in the
Worthington community. In the first passage above, he attaches strong import-
ance to the city’s political boundary. In the second passage, however, he
explains that his continued involvement in Worthington activities allowed
him to overcome the alienation from the community that could have resulted
from living outside the city boundary. Still, it is clear that Greg perceives the
city boundary as marking a real social division involving loyalty to the
Worthington community.
It is this type of perception that another speaker, Don, acknowledges but
quickly dismisses. Don lives just outside the Worthington city boundary and
has served on the Worthington Board of Education.
6. Don: When the school boundaries were set way back . . . and set, fixed
essentially for all time, which took in a lot of Columbus, you know . . . it
affected I think people in the city itself that weren’t, felt like people
lived outside of the city boundary were a little bit carpet-bagging on the
good name of Worthington, you know.
...
But I think basically I think that’s a very misguided view, because I
think the people that do live [in the outskirts] have a very strong
affinity and connection to the town. So it’s a misunderstood and
misplaced view of, of, of, you know, about people who live here
because I really believe many people live outside the boundaries of the
real city of Worthington carry just as strong an affinity and feeling for,
and support for the town itself, as those that live in the town . . . I mean I
feel like that’s where I live. I don’t feel like I live in Columbus or
close to Dublin or or or just south of Powell or in Sharon Township or
you know or whatever else, I, I feel like I’m from Worthington. I
don’t even stop to think about where it is. I rarely ever think of the
boundaries.

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 247

Although he claims that it has real importance only for people who live in the
city of Worthington, Don is clearly conscious of the city boundary. For him,
the critical factors are (or should be) more subjective: affinity and support for
Worthington take precedence over location. In this respect, Don’s view resem-
bles Rita’s insofar as Rita also cites subjective indicators of community mem-
bership such as feeling connected. However, she explicitly excludes West
Worthington, where Don lives, from her definition of Worthington. Don’s insis-
tence on denying the importance of the city boundary and location generally
is not even shared by certain residents of the outskirts, including Dana, who
attends church and shops in Old Worthington but lives considerably further
outside the city than Don does.
7. Dana: I feel like I’m from Columbus, I live in Columbus . . . I mean I
do activities in Worthington, but, um, I’m a little bit of an outsider I
suppose. You know, cause I can’t, cause I’m not in Worthington
proper, and I don’t have access to some Worthington activities . . . like
I’m in Worthington schools but I can’t get a Worthington membership
to the community center, I have to pay the extra surcharge . . . [Dana’s
neighborhood is] kind of more of a suburb, like, a sprawl of Columbus
I suppose . . . I think that people that live in Worthington feel very tied
to the city limits of Worthington, but I’m not really in the city limits,
so I can’t really call myself somebody from Worthington I guess . . . I
think there’s a sense of community that you find in a more established,
defined area that I don’t know that I find in community sprawl out
here.
Dana embodies the disconnected resident that Rita imagines: living in an
anonymous, boundary-free neighborhood that she describes as ‘community
sprawl,’ Dana does not feel like a true member of the Worthington community
and even experiences her outsider status in material ways like her exclusion
from Worthington’s recreation center. A dominant claim is that people living
outside the city boundaries have little involvement in the Worthington commu-
nity beyond school activities, shopping, or attending church; they are, there-
fore, not ‘real’ community members. Definitions of local space underlie status
in the community (cf. Modan 2002).
Comparing particular individuals sheds some light on the observed linguis-
tic patterns. Rita and Dana graduated from the same high school at roughly
the same time, currently stay home with their young children, are married to
professionals, earned graduate degrees in the same field from the same school,
and think highly of one another. Recall, however, that they live in different
regions of Worthington: Rita lives at the heart of Old Worthington and Dana
lives well into the outskirts. As the passages from their interviews above indi-
cate, Rita identifies strongly with the Worthington community and views it as
threatened, while Dana considers herself an outsider and shows no evidence
of viewing the community’s cohesion as endangered. Thus their self-perceived
positions with respect to the Worthington community are essentially

# Blackwell Publishing Ltd. 2005

248 DODSWORTH

opposites. Further, Rita has had the most community involvement of all the
speakers, while Dana only attends church and shops in Worthington. The two
speakers also exhibit markedly different rates of /l/ vocalization. As shown in
Table 7, Rita, who has the most community involvement and the strongest feel-
ings about Worthington’s (threatened) identity, has the lowest rate of /l/ vocal-
ization. Dana, who considers herself an outsider, has a much higher rate.
These facts suggest that resistance to linguistic changes in progress is a com-
ponent in the effort to maintainWorthington’s identity as a community distinct
from Columbus, an increasingly difficult task as Columbus develops the land
surrounding Worthington. Thus the patterns of linguistic variation presented
here are products of, and tools in, the construction of a range of stances regard-
ing Worthington’s community identity in the face of urban sprawl. Many
Worthington residents, such as Rita, value the city’s social and geographic dis-
tinctiveness, while many who live outside the city benefit from the physical
and cultural integration of the surrounding areas with Worthington. By using
the older linguistic variants, speakers like Rita resist the disappearance of the
Worthington community into Columbus sprawl. These speakers’ use of the lin-
guistic variables is a component in the construction of their personae as true
‘Worthingtonites’ (a local term indexing loyalty to the city). Of course, this
claim cannot be confirmed without a matched guise or other type of percep-
tual test.
The 2003 Worthington bicentennial events illustrate the connection
between Worthington identity and attachment to continuity, or resistance to
change. These events were planned over three years by a committee made up
of volunteers (including Rita and Greg) who live in the city, many of them
known for their active engagement with city organizations and events. Most of
the bicentennial events re-enacted some aspect of Worthington’s early culture.
For instance, at a formal ball held as a fundraiser, participants were encour-
aged to wear elegant 19th century costumes. There was also a locally written
and produced play featuring Worthington’s founding settlers as characters.
These events and others foregrounded and idealized an earlier era when
Worthington, with its New England-style culture, was culturally and geogra-
phically distinct relative to Columbus. The events, therefore, implicitly con-
demn the loss of Worthington’s independence and work to preserve it as the
city becomes integrated with the surrounding, newly developed areas of
Columbus.

Table 7: Two speakers’ rates of /l/ vocalization

Speaker Percentage /l/ vocalization

Rita 10.8 (4/37)

Dana 20.0 (7/35)

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 249

One of the final bicentennial events, held in September 2003, was called
StreetFest and included dozens of musical performances, activity booths for
children, and food vendors set up on the main street running through Old
Worthington. From an ethnographic perspective, StreetFest was interesting
for several reasons. One reason is that the children’s activities revolved around
the theme of international diversity ^ each of the activity booths, for instance,
offered a craft from a different country. Although several speakers mentioned
that Worthington has become a much more diverse community in recent
decades, the diversity theme at StreetFest appeared to be confined to interna-
tionalism, rather than any kind of locally relevant diversity. None of the events,
for instance, celebrated the broadened socio-economic and ethnic diversity of
students who benefit from the undeniably superb Worthington school system.
Another ethnographically interesting element of StreetFest was a large plas-
tic board, perhaps seven feet high and three feet wide, with a title at the top
that read ‘Worthington’s Next 200 Years: Building the Future.’ At a nearby
table, people passing by could write messages on self-adhesive slips of paper
and place them on the board. By the end of the day, hundreds of messages com-
pletely covered the board. Most of the messages did not address Worthington’s
future, as the title suggested, but focused instead on perceptions and feelings
about the community. Three are copied in Example 8:
8. a. I like the way it stays the same.
b. Like it was 50 years ago.
c. I hope the library is still a vital part of Worthington in 200 years. It
would be a tribute to our founders to value lifelong learning in the
future as much as we do today, and as much as our founders did in
1803!
Messages a and b suggest that Worthington is perceived by its own resi-
dents as an unchanging community, and that, at least for the author of
message a, this is a positive quality. It is unlikely that the people who
wrote these two messages are unaware of the changes that the community
has undergone in recent years: nearly all the speakers older than 25 noted
the ongoing development of northern and western areas both inside and outside
the city boundary, and many speakers also discussed the controversial
construction of a second high school during the early 1990s. Instead, the two
writers probably recognize important changes but also perceive the purpose-
ful maintenance of Worthington’s colonial, small-town atmosphere. The
bicentennial events largely highlighted this aspect of Worthington,
strengthening its role in the city’s self-perceived identity. Message c points to
the importance of collective beliefs about the founding settlers and the values
they espoused; more than one speaker attributed the quality of Worthington’s
school system to the founders’early emphasis on education. All three messages
reveal the perception that there has been some kind of desirable continuity
throughout Worthington’s history. This attachment to sameness is consistent

# Blackwell Publishing Ltd. 2005

250 DODSWORTH

with the view that Worthington’s identity and its distinctive community must
be protected from the ongoing urban growth.
The patterns of linguistic variation, then, reveal, and are part of, a division
between two broad groups: on one hand, people who want to preserve a certain
amount of exclusivity and closeness in the Worthington community; and on
the other hand, people who recognize this desire but either do not take part
in it or actively dismiss it. Within each of the two broad groups, of course,
there is variation both in use of the linguistic variables and in ideology.
Speakers living in Old Worthington vocalized /l/ less frequently but accord-
ing to the same linguistic pattern as speakers living in the city but outside
Old Worthington. While both sets of speakers generally appreciate the city’s
New England small-town identity, the colonial atmosphere is more salient
on a daily basis to those in Old Worthington, who live within sight of the
landmarks that lend the town its colonial feel. There is also ideological
variation among Old Worthington residents; some of them resist any efforts
to exclude Columbus residents from the Worthington community. There
exists similar ideological variation among Columbus residents, as between
Don and Dana. Accordingly, the use of the linguistic variables differs across
individuals as they construct their own styles (cf. Eckert 2002).

CONCLUSIONS AND FUTURE WORK

One area left unexplored here is the technique’s potential for modeling the con-
nections between local and supra-local social structures (cf. Eckert 2000;
Meyerhoff 2002), or between abstract social constructs such as class and local
instantiations of them. The social parameters used in attribute networking are
locally recognized, a trait that much work (e.g. Labov 1963; Le Page and
Tabouret-Keller 1985; Rickford 1986) has identified as critical, but they will
resist comparison across communities unless they can be linked to non-local
social constructs. Along those lines, an area ripe for analysis in the
Worthington aggregate network is the relationship between wealth and local
geopolitical boundaries. As shown in Table 8, three levels of wealth emerge
from the network, and each level is linked to several geographic areas. We see
not only that class is locally perceived as instantiated by three broad levels of
wealth, but also that local geography plays an important role in defining the
three‘classes.’ For class and geography to intersect meaningfully is predictable;
but an analysis that aims to consider multiple perceptions of geographic space,
economic stratification, and community identity simultaneously can surely
benefit from a systematic technique for representing and unifying speakers’
perceptions, especially if it is to be useful for cross-community comparison.
One of the advantages of attribute networking as a quantitative technique is
that attributes having social significance for only a few people are not likely to
be selected by the criteria for structural importance. Conversely, attributes

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 251

Table 8: Levels of wealth and associated geographic spaces in the aggregate

network
Geographic areas directly
Attributes used to connected to the attributes in the
Level of wealth identify the level of wealth middle column

High upscale community W proper

expensive homes Old W,W Hills
wealthy neighborhood Riverlea, west of Olentangy,W Hills
highest socio-economic level W Hills
really rich people W Hills
Moderate moderate houses W Estates
have money but not really rich east of Old W
middle economic status Kilbourne region
Low low socio-economic level Colonial Hills, Old W
struggle financially east part of Colonial Hills
apartments north-west of SR-161, around Selby,
outskirts

that have broad social significance are likely to be identified as structurally

important even though all informants are expected to directly express only a
fraction of their conception of the community’s social space during ethno-
graphic interviews. Therefore, the technique is likely to identify social issues
that have broadly felt importance in the community. There may, however, be a
certain minimum number of informants, relative to the community’s size,
required to ensure representation of important social perceptions.
In conclusion, attribute networking is put forth as a type of technique that,
combined with existing approaches, could advance the study of subjective
experiences and their interaction with sociolinguistic variation. Clearly this
paper does not exhaust the technique’s potential; for instance, no attempt has
been made here to form categories out of densely connected sets of nodes. I
hope, however, that the usefulness of this kind of technique for the study of
sociolinguistic variation has been clearly demonstrated.

NOTES
1. Many thanks especially to Don Winford for commenting on numerous drafts and
for his continuing encouragement. Thanks also to others whose suggestions
have greatly improved the paper: Jim Moody, Galey Modan, Keith Johnson, Mary
Beckman, Scott Schwenter, and two anonymous reviewers.
2. Some of the apparent disagreement in the aggregate network undoubtedly results
from the variety of ways that speakers stated their perceptions, and the intercalation
process may have intensified it: two similar attributes, from two separate interviews,

# Blackwell Publishing Ltd. 2005

252 DODSWORTH

were not treated as identical unless the discourse contexts left no doubt that they
shared a referent. This strictness was probably unnecessary in certain cases; some
of the similar nodes could probably be collapsed, creating higher-valued ties.
However, merging nodes that are similar but not identical would result in a loss of
nuance and confidence.
3. Polaris is the site of a large, upscale shopping mall north of Worthington.

REFERENCES
Ash, Sherry. 1982a. The vocalization of intervocalic /l/ in Philadelphia. SECOL Review
6: 162^175.
Ash, Sherry. 1982b. The vocalization of /l/ in Philadelphia. Ann Arbor, Michigan:
University Microfilms International.
Allen, Harold. 1976. Linguistic Atlas of the Upper Midwest (volume 3). Minneapolis,
Minnesota: University of Minnesota Press.
Bearman, Peter S., Robert Faris and James Moody. 1999. Blocking the future: New solu-
tions for old problems in historical social science. Social Science History 23: 501^533.
Borgatti, Stephen P. and Martin G. Everett. 1999. Models of core/periphery structures.
Social Networks 21: 375^395.
Borgatti, Stephen P., Martin G. Everett and Linton C. Freeman. 2002. Ucinet for
Windows: Software for social network analysis. Harvard: Analytic Technologies.
Brooks, Candy. 2004. Worthington ARB faces new challenges in a changing economy.
This Week in Worthington, April 1: A4.
Carter, Bob and Alison Sealey. 2000. Language, structure and agency: What can
realist social theory offer to sociolinguistics? Journal of Sociolinguistics 4: 3^20.
Carver, Craig. 1993. Regional speech variation in Wisconsin. In Timothy Frazer (ed.)
‘Heartland’ English: Variation and Transition in the American Midwest. Tuscaloosa,
Alabama and London: University of Alabama Press. 199^214.
Durian, David. 2004. /s/ retraction and /l/ vocalization realization in Columbus, Ohio,
AAVE speech: A sociophonetic analysis. Unpublished manuscript. Columbus, Ohio:
The Ohio State University.
Eckert, Penelope. 2000. Linguistic Variation as Social Practice: The Linguistic
Construction of Identity in Belten High. Malden, Massachusetts: Blackwell.
Eckert, Penelope. 2002. Constructing meaning in sociolinguistic variation. Paper pre-
sented at the Annual Meeting of the American Anthropological Association, New
Orleans, Louisiana, November.
Fairclough, Norman. 2000. Response to Carter and Sealey. Journal of Sociolinguistics 4:
25^29.
Freeman, Linton C. 1977. A set of measures of centrality based on betweenness.
Sociometry 40: 35^41.
Hardcastle,William and William Barry. 1985. Articulatory and perceptual factors in /l/
vocalizations in English. Journal of the International Phonetic Association 15: 3^17.
Hazen, Kirk. 2002. Identity and language variation in a rural community. Language 78:
240^257.
Horvath, Barbara M. and Robert J. Horvath. 2002. The geolinguistics of /l/ vocalization
in Australia and New Zealand. Journal of Sociolinguistics 6: 319^346.
Labov, William. 1963. The social motivation of a sound change.Word 19: 273^309.

# Blackwell Publishing Ltd. 2005

ATTRIBUTE NETWORKING 253

Labov, William. 2001. Principles of Linguistic Change, Volume 2: Social Factors. Oxford,
U.K.: Blackwell.
Labov,William, Paul Cohen, Clarence Robins and John Lewis. 1968. A Study of the Non-
standard English of Negro and Puerto Rican Speakers in New York City (Cooperative
Research Report 3288) Volume I: Phonological and Grammatical Analysis. Philadelphia,
Pennsylvania: U.S. Regional Survey.
Lane, Lisa Ann. 1998. Emergence and transformation of a dialect:Thybor
nsk (Danish).
Unpublished dissertation. Chicago, Illinois: University of Chicago.
Le Page, Robert and Andre¤ e Tabouret-Keller. 1985. Acts of identity: Creole-based
approaches to language and ethnicity. Cambridge, U.K.: Cambridge University Press.
McCormick, Robert and Jennie McCormick. 1992.Worthington Landmarks: Photo-essays
of Historic Worthington Properties. Worthington, Ohio: Cottonwood Publications.
Mendoza-Denton, Norma Catalina. 1997. Chicana/Mexican identity and linguistic var-
iation: An ethnographic and sociolinguistic study of gang affiliation in an urban
high school. Unpublished PhD dissertation. Stanford, California.: Stanford University.
Meyerhoff, Miriam. 2002. Communities of practice. In J. K. Chambers, Peter Trudgill
and Natalie Schilling-Estes (eds.) The Handbook of Language Variation and Change.
Malden, Massachusetts: Blackwell. 526^548.
Mills, C. Wright. 1959. The Sociological Imagination. London: Oxford University Press.
Milroy, Lesley. 1980. Language and Social Networks. Oxford, U.K.: Blackwell.
Milroy, Lesley. 1987. Language and Social Networks (2nd edition). Oxford, U.K.: Blackwell.
Milroy, Lesley and James Milroy. 1992. Social network and social class: Toward an inte-
grated sociolinguistic model. Language in Society 21: 1^26.
Modan, Gabriella. 2002. Public toilets for a diverse neighborhood: Spatial purification
practices in community development discourse. Journal of Sociolinguistics 6: 487^513.
Potter, Jonathon. 2000. Realism and sociolinguistics. Journal of Sociolinguistics 4: 21^24.
Rickford, John. R. 1986. The need for new approaches to social class analysis in sociolin-
guistics. Language and Communication 6: 215^221.
Sproat, Richard and Osamu Fujimura. 1993. Allophonic variation in English /l/ and its
implications for phonetic implementation. Journal of Phonetics 21: 291^312.
U.S. Census 2000. http://factfinder.census.gov
Wasserman, Stanley and Katherine Faust. 1994. Social Network Analysis: Methods and
Applications. Cambridge, U.K.: Cambridge University Press.
Wells, John C. 1982. Accents of English. Cambridge, U.K.: Cambridge University Press.

Address correspondence to:

Robin Dodsworth
Linguistics Department
222 Oxley Hall
1712 Neil Avenue
Columbus
Ohio 43210^1298
U.S.A.
[email protected]

# Blackwell Publishing Ltd. 2005