Language variation and change in social networks

A bipartite approach

Robin Dodsworth, Richard A. Benton

• Publisher: Routledge
• Year: 2019
• ISBN:
  9781317281702
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Linguistics
• Index: Languages & Linguistics

This monograph takes up recent advances in social network methods in sociology, together with data on economic segregation, in order to build a quantitative analysis of the class and network effects implicated in vowel change in a Southern American city.

Studies of sociolinguistic variation in urban spaces have uncovered durable patterns of linguistic difference, such as the maintenance of blue collar/white collar distinctions in the case of stable linguistic variables. But the underlying interactional origins of these patterns, and the interactional reasons for their durability, are not well understood, due in part to the near-absence of large-scale network investigation. This book undertakes a sociolinguistic network analysis of data from the Raleigh corpus, a set of conversational interviews collected form natives of Raleigh, North Carolina, from 2008-2017. Acoustic analysis of the corpus shows the rapid, ongoing retreat from the Southern Vowel Shift and increasing participation in national vowel changes. The social distribution of these trends is explored via standard social factors such as occupation as well as innovative network variables, including a measure of nestedness in the community network.

The book aims to pursue new network-based questions about sociolinguistic variation that can be applied to other corpora, making this key reading for students and researchers in sociolinguistics and historical linguistics as well as those interested in further understanding how existing quantitative network methods from sociological research might be applied to sociolinguistic data.

1
Previous Approaches to Network Analysis in Sociolinguistics

Decades before linguists were systematically studying the relationship between language variation and social structures, Leonard Bloomfield offered a network-based explanation for local and regional isoglosses:

The reason for this intense local differentiation is evidently to be sought in the principle of density…. Every speaker is constantly adapting his speech-habits to those of his interlocutors; he gives up forms he has been using, adopts new ones, and, perhaps oftenest of all, changes the frequency of speech-forms without entirely abandoning any old ones or accepting any that are really new to him. The inhabitants of a settlement, village, or town, however, talk much more to each other than to persons who live elsewhere. When any innovation in the way of speaking spreads over a district, the limit of this spread is sure to be along some line of weakness in the network of oral communication, and these lines of weakness, in so far as they are topographical lines are the boundaries between towns, villages, and settlements.

(1933: 326–8)

Bloomfield postulated that linguistic diversity was promoted not only by geographically determined “lines of weakness” in communication between populations, but also by socially determined lines of weakness within communities (1933: 47). Labov (2001: 228) restates Bloomfield’s theory as “the more often people talk to each other, the more similar their speech will be.” The broad perspective that social network structure affects linguistic variation has long been influential in sociolinguistics. Even if the majority of variationist sociolinguistic studies do not include direct information about speakers’ social networks, their hypotheses and conclusions are usually consistent with the idea that people in close contact will talk alike.

In most sociolinguistic studies that explicitly investigate social network effects, the network data are ego-centric, having to do with the characteristics of individual speakers’ first-order networks rather than with the community’s overall network structure (Sharma 2017). Our study of linguistic variation in Raleigh, North Carolina, is unusual in this respect. We represent speakers’ positions in the overall community network using a bipartite approach described in Chapter 3. For that reason, our quantitative methods and conclusions differ in some respects from those of previous sociolinguistic research, but our hypotheses nevertheless build directly upon well-established sociolinguistic theory and evidence concerning language and social networks.

This chapter summarizes sociolinguistic studies from the past 50 years involving social network data in some form. The chapter is structured as a series of common guiding principles, each illustrated with detailed examples. We conclude with an appraisal of the benefits and limitations of the network analysis in these studies, setting the stage for explaining how our study of linguistic variation and social network structure in Raleigh both builds upon and diverges from previous work.

1.1 Guiding Principles and Their Realization in Previous Studies

1.1.1 Social Network Is a Better Unit of Analysis Than Speech Community

Many early studies explicitly used social networks as a replacement for the Labovian speech community, in which socioeconomic class, age, and sex are typically the axes of social variation. While the shift away from this top-down view of communities was part of a more general theoretical turn in sociolinguistics during the 1980s and 1990s, it was also motivated by the social and linguistic details of the populations being investigated. In particular, sociolinguists turned to social networks as the relevant units of social structure in communities where social classes, in Labov’s sense of hierarchically ordered groups, either were difficult to define or did not exist. There were several ways in which the Labovian social class framework proved problematic in some communities.

First, in contrast with the largest first-wave sociolinguistic studies of urban areas (e.g., Labov 1966, 2001; Trudgill 1974; Wolfram 1969) that used large speaker samples with wide-ranging economic characteristics, studies of smaller communities or of single neighborhoods tend to involve groups of economically similar speakers. For example, Milroy (1987) investigates three working class neighborhoods in Belfast where male unemployment is high and the population generally is “blighted” and suffers from “social malaise” (72). Milroy reports that all of the speakers in the Belfast sample are comparable to the lower working class in Trudgill’s (1974) Norwich sample. Similarly, Edwards (1992) studies speakers in an impoverished neighborhood in southeast Detroit; while some areas of the neighborhood are more affluent than others, the overall rate of unemployment is high, and most employed residents hold “lower status positions” in service, manufacturing, or as laborers (97). In communities such as these, all of the speakers would be classified at one end of the scale in a first-wave sociolinguistic class model. So the social network model was adopted not only as a way of showing social differentiation within economically homogeneous communities, but also as a framework within which to develop hypotheses about linguistic variation within relatively small groups.

Another reason for replacing the traditional speech community framework with a social network model was that local interactional factors were seen as more important than economic differences in the context of daily life. For example, Lippi-Green (1989) reports that in the dairy farming village of Grossdorf, Austria,

[I]t is not occupation that determines one’s place in the hierarchy, but rather the degree of integration into the established structures. A successful farmer from a well-established clan may not make as much money or build as nice a home as someone well-placed in the provincial government, but it is the farmer who is more likely to be elected to the Council. It is not so much a matter of class or status, but who you know, and who knows you.

(Lippi-Green 1989: 216)

Milroy (1987: 12–17) makes the related point that people in many working class communities are first and foremost locally oriented, having low levels of geographic and social mobility and close relationships with neighbors. In the course of ethnographic observation in Belfast, Milroy finds that community members prioritize local community cohesion:

Most important for our analysis here is the very high value placed on social solidarity. In all three communities this was commonly valued above everything else, including improved material conditions; the maintenance of solidary relationships was seen as a necessary buffer in times of sickness or need, or against hostile outsiders, particularly the authorities. These values were often expressed in terms of being “neighbourly” or “looking after your own”.

(Milroy 1987: 73)

Thus in some types of communities, linguistic variation has more to do with neighborhood residence, and neighborhood status, than with economic characteristics. In fact, Milroy (1987: 157) explicitly proposes that linguistic variants can function as network markers. In other words, linguistic variables pattern by network not just because people in the same dense, multiplex network are exposed to many of the same linguistic forms, but also because they actively use those forms to index belonging to a localized network. A network model is, in theory, better suited than a class model for uncovering identity-driven sociolinguistic differences in these communities.

A third reason for replacing social class with social network in sociolinguistic studies is that in some cases, the speakers are children. Although classifying children according to their parents’ economic situations does give some indication as to their lived experiences, there are several clear shortcomings to this approach. The most important is that in the course of their daily lives, most children interact with, and learn linguistic norms from, their peers, regardless of what their parents are doing. Several socio-linguistic studies have therefore used network data to approximate patterns of interaction in child and adolescent friendship groups, especially in studies of linguistic variables associated with ethnic groups (Cheshire et al. 2008; Fox 2007; Khan 2006; Labov 1972; Labov & Harris 1986; Nardy, Chevrot, & Barbu 2014).

A fourth reason for avoiding the traditional speech community framework is that the population lacks a unified set of linguist norms. Labov’s (1972) definition of speech community hinged on the existence of a shared set of evaluative norms as well as shared patterns of linguistic variation with respect to internal and stylistic factors. Many sociolinguists, however, have observed that the social meanings associated with linguistic variants can differ from one group to the next within a community. A related problem is that, in contrast with most first-wave studies, the community may lack a single, well-defined language or dialect. Le Page and Tabouret-Keller (1985) deal with an extreme version of this situation, investigating variability among speakers of Caribbean English Creoles. The linguistic and social variability in the sample was too multi-dimensional for a hierarchical class model, or any kind of unified community model, to be viable.

Finally, many sociolinguists who adopt network approaches have done so out of the desire to study linguistic variation from person to person, rather than from group to group (e.g., working class to middle class, or male to female). On one hand, this is a question of statistical strategy. Milroy (1987) gives particular emphasis to this point (e.g., pages 149–152). In first-wave approaches, individual speakers are often invisible because linguistic data are calculated as group-level aggregates. By contrast, in some social network approaches, rank-order correlations are calculated from individual speakers’ linguistic and network data (e.g., Milroy 1987; Bortoni-Ricardo 1985). On the other hand, the issue of the individual-versus-group has to do with the fact that network studies represent something about each individual’s interactional patterns or position within the local community, instead of (or in addition to) grouping speakers into class and gender categories.

All of the aforementioned arguments notwithstanding, social network data are often understood as a complement to traditional aggregate data, not as a replacement (Labov 2001; Milroy & Milroy 1992). For example, network data have the potential to reveal the interactional causes of the aggregate class patterns that have appeared and reappeared in first-wave sociolinguistic studies. Likewise, the Raleigh study described in this book yields the conclusion that network approaches and hierarchical class approaches complement one another.

1.1.2 Community Integration Correlates With Use of Local Linguistic Variants

A central hypothesis in social network studies in sociolinguistics has been that speakers’ level of integration into the local community network correlates positively with their use of local linguistic variants. There are effectively three reasons. First, membership in a local community, especially one in which many people know one another (a densely connected network), means frequent exposure to the community’s linguistic norms simply because members are likely to interact with one another often. Second, membership in a densely connected community can also mean isolation from other communities, and therefore lack of exposure to other linguistic norms (Centola 2015; McPherson, Smith-Lovin, & Cook 2001). Dense, multiplex networks act as norm enforcement mechanisms, as articulated by Bott (1957) (see also Centola, Willer, & Macy 2005; Friedkin 1993; Moody & White 2003). Third, interaction with other community members brings about not only exposure to linguistic norms, but also the evaluation of one’s own speech by others. Just as local linguistic variants often carry positive indexicality such as friendliness or toughness, deviation from local norms invites negative judgement. Both for reasons of exposure and for reasons of evaluation, speakers who are well-integrated into tightly connected local communities are theorized (and sometimes observed) to adhere to local norms.

Harlem (Labov 1972)

Labov’s (1972) study of Harlem, a neighborhood in New York City, is an early example of assessing vernacularity as a function of integration in the local community. Labov investigates the relationship between peer group membership and use of core features of African American English (AAE) among the Thunderbirds (T-Birds), a social group of adolescent and pre-adolescent boys in a single apartment building. In the same apartment building, Labov also considers four non-members (“Lames”), all age 10. The Aces are a social group of boys from another apartment building. As a comparison group, there is additionally a ten-speaker sample from the larger corpus of Vacation Day Camp speakers, who are a mix of members and non-members spanning a wider geographic region. All of the boys belong to working class families.

The initial linguistic question is whether membership in one of the social groups (T-Birds or Aces) promotes the use of the local vernacular. If so, then the Lames are expected to use lower rates of AAE features than group members. This turns out to be true for some variables. For example, the Lames show the highest rates of coda /r/ presence in all three of the linguistic styles (conversation, reading passage, and word lists together with minimal pairs). The Lames also show a sharper upward slope from interview to reading passage to word list. Additionally, the Lames also show lower rates of (dh) stopping and affrication across all styles, except that the Vacation Day Camp boys have a lower rate in the reading passage style.

The “most sensitive sociolinguistic variable” for AAE speakers, (ing), shows a different pattern (Labov 1972: 266). All the boys show near categorical [iŋ] during the reading passage and word list styles. In the conversational style, however, the Aces and T-Birds produce [iŋ] at rates at or near 0, in contrast with rates of 24 and 22 for the Vacation Day Camp boys and the Lames, respectively.

A particularly complex variable, with respect to the relationship between internal factors and social group membership, is (t/d) deletion. The two major internal factors governing this variable are the morphological status of the /t/ or /d/ and the nature of the following sound. In interview speech only, all of the groups delete (t/d) at rates in the 90s when there’s a following consonant in monomorphemic contexts, the context most favoring deletion. When there’s a following vowel, the Aces, T-Birds, and Lames drop to deletion rates around 60, and the Vacation Day Camp speakers drop to 35. In bimorphemic contexts, a following consonant is associated with high deletion rates for the Aces (81), T-Birds (74), and Vacation Day Camp speakers (81), but a rate of only 19 for the Lames. Finally, a following vowel in bimorphemic contexts means low deletion rates all around (24 each for the T-Birds, Aces, and Vacation Day Camp boys, 16 for the Lames). The conclusion to be drawn from these results is that for the members of the Aces and T-Birds, phonological context is the stronger constraint: a following vowel disfavors deletion. By contrast, for the Lames, as for many White speakers in New York City and elsewhere (cf. Guy 1980), morphological status is the stronger constraint: the past tense morpheme disfavors deletion to a greater degree than a following vowel. Labov attributes the difference to the Lames’ greater influence from the local White vernacular and lesser exposure to the local AAE dialect. By virtue of their lesser integration in the local adolescent community, the Lames not only use lower rates of (t/d) deletion, but also (remarkably, in our view) acquired a different ranking of internal constraints.

Another core AAE feature is copula contraction and deletion. The data (while sparse for the Lames) indicate that although the Lames contract at rates similar to the T-Birds, they rarely delete the copula. By contrast, the T-Birds’ deletion pattern reflects the expected constraints of subject type (NP vs. pronoun) and the following grammatical category.

Other linguistic differences between the T-Birds and Lames include: a higher rate of it as dummy subject by T-Birds; a higher (almost categorical) rate of negative concord by T-Birds; and a higher rate of inverted word order in embedded questions by T-Birds. In summary, the members of the local peer groups, the T-Birds and the Aces, show not only higher rates of local vernacular features, but also different patterns with respect to internal constraints, relative to the Lames.

An expanded view of the role of peer group membership (pages 271–273) compares 31 members of the T-Birds, Aces, Jets, and Cobras collapsed into one group, with the (older) Oscar Brothers group, a group of ten Lames from the T-Bird, Cobra, and Jet areas, and eight White boys. The question is to what extent each group shows subject-verb agreement with have, do, don’t/doesn’t, want, and say. As expected, the members of the peer groups show the lowest rates of agreement, followed by the Lames, followed by the White speakers. The same pattern emerges for frequencies of past tense were. This is further evidence of the linguistic influence of integration into a local peer group.

A separate network-based look at the linguistic consequences of peer group membership involves the Jets, one of the local social groups. Thirty-six boys, all members or would-be members of the Jets, are classified as core, secondary, peripheral, and Lame members on the basis of observation as well as friendship naming by the boys. A further division is made between members found on the 100s block vs. the 200s block, as this geographic distinction corresponded to different hangout groups. Network status appears to have linguistic consequences for copula deletion: core members delete with the highest rates, followed by secondary members, and then peripheral and Lame members (Table 7.8, page 279). Other linguistic variables, however, do not clearly pattern with network position, including contraction and also clause-internal negative concord. Some other features, including the tag an’ shit, are used mainly by core members.

Belfast (Milroy 1987)

Milroy’s (1987) sociolinguistic network study of three neighborhoods i...

Table of contents

• Cover
• Half Title
• Series Page
• Title
• Copyright
• Contents
• List of Figures
• List of Tables
• Acknowledgments
• 1 Previous Approaches to Network Analysis in Sociolinguistics
• 2 Raleigh, the Corpus, and the Retreat From the Southern Vowel Shift
• 3 Bipartite Networks and Complex Social Systems
• 4 Structural Equivalence
• 5 Community Detection
• 6 Conclusions
• Appendix 1 Definitions of Selected Terms
• Appendix 2 Full Model Summaries from Chapter 2, Section 2.5
• References
• Index