Why some behaviors spread while others don’t: A laboratory simulation of dialect contact

Abstract

The question of how behavioral variants compete and propagate is of primary importance to the study of cultural evolution; with respect to language, it is also an important focus of the field of sociolinguistics. Variant propagation can occur by neutral means—akin to drift in biological evolution—or through selection, whereby individuals are biased in what variants they adopt. An important bias concerns social meaning, and sociolinguistic theory distinguishes between variants that are primarily associated with a particular social group (such as working-class people or Texans) and variants primarily associated with a perceived trait of the group (such as toughness). In the former case, variants are hypothesized to propagate neutrally; in the latter case, provided the trait is socially relevant to adopters, variants are hypothesized to be subject to selection and to propagate more readily. To test this hypothesis we conducted an experimental study in which groups of four participants played a game that involved instant messaging in an artificial “alien language” with two dialects. Each player was assigned to one of two alien species, the weaker Wiwos or the tougher Burls. The social meaning of one feature of the Burl dialect was manipulated, and the results strongly supported the hypothesis: Variants from the Burl dialect were used by Wiwos in all conditions, but when associated primarily with “tougher aliens”, the rate of adoption was significantly greater than when they were associated primarily with “Burls”. When toughness was made irrelevant to the game, the effect of association disappeared, emphasizing the importance of social relevance in the propagation of behavioral variants.

Introduction

This paper presents an experimental study of how social factors influence the cultural evolutionary spread of linguistic variants. Cultural evolution can be defined as change in the frequency of cultural variants over time. Examples of cultural variants range from the particular design of a Persian rug (Matthews, Tehrani, Jordan, Collard, & Nunn, 2011), arrowhead (Mesoudi & O’Brien, 2008), or folktale (da Silva & Tehrani, 2016) to gait velocity (Ebersbach, Sojer, Müller, Heijmenberg, & Poewe, 2000) or the features of a speech sound (Croft, 2000, Ritt, 2004). An important aspect of research on cultural evolution is understanding the way in which cultural variants propagate between individuals and populations, often outcompeting and replacing other variants. Linguistic cultural variants are particularly good candidates for investigation for two reasons. First, they are ubiquitous: All human societies have language, and all members of those societies (barring pathological cases) use language daily, making it a rich source of data. Second, languages are constantly changing, and this process of language change has been the object of study in one form or another for several centuries, leading to a relative abundance of data and theoretical work.

Broadly speaking, there are two fields for which the propagation of linguistic variants is an important focus of study. The first, sociolinguistics, emerged as a distinct field in the 1960s, and is primarily concerned with investigating real-time variation and change in modern language, including change brought about by social factors (Labov, 2001), but also by language-internal (Labov, 1994), cognitive, and cultural factors (Labov, 2010). The second field, language evolution, coalesced in the last decades of the twentieth century. The field focuses not only on the origins and biological evolution of the language faculty, but also on the study of language change as a cultural evolutionary process (Croft, 2000, Ritt, 2004), with increasing attention being paid to the role that cultural evolution plays in the origins of complex language (Dediu et al., 2013, Thompson et al., 2016).

Standard methods in sociolinguistics involve a combination of ethnographic research and interviews designed to elicit naturalistic and vernacular speech (Labov, 1984). Sociolinguistic interview data and other speech samples have been compiled into large-scale speech corpora, such as the Philadelphia Neighborhood Corpus (University of Pennsylvania, 1972-2013), which provide a broad view of variation across a speech community over time. The field has also employed experimental methods since its early days, and in recent years these have become increasingly common (e.g., Campbell-Kibler, 2010, Lambert et al., 1960, Preston, 2010), though ethnographic approaches continue to dominate. By contrast, empirical work in language evolution has tended to rely heavily on computer simulations and controlled laboratory experiments, with more minor or indirect reliance on real-world sociolinguistic data. Experiments in this field often involve the use or collaborative development by participants of novel communication systems under controlled conditions, a line of research that has been termed Experimental Semiotics (see Galantucci, Garrod, & Roberts, 2012, for a review); while some studies of this kind have investigated questions very relevant to sociolinguistics (e.g., Kerr and Smith, 2016, Roberts, 2010), such methods have very little foothold in that field.¹

In this paper we present an Experimental Semiotic study designed to directly test a hypothesis drawn from sociolinguistic theory and inspired by sociolinguistic fieldwork. The experiments will be presented in Sections 2 Experiment 1: Tough Talk vs. Burl Talk, 3 Experiment 2: No fighting. In Section 1.1 we discuss the hypothesis to be tested and its sociolinguistic background.

From a cultural evolutionary point of view, the propagation of variants in a population may be governed either by neutral forces (such as sampling error) or by selection, in which speakers are biased in some way. Richerson and Boyd (2006, p. 69) categorized cultural evolutionary biases into three main types: content-based bias, based on the nature of a variant (such as ease of pronunciation or comprehension), frequency-based bias, based on its commonness or rarity,² and model-based bias, based on the social identity or characteristics of individuals bearing the variant. Although these terms are typically not used in sociolinguistics, all three kinds of bias have received attention in that field, with model-based bias attracting particular attention (e.g., Bucholtz, 1999, Eckert, 2000, Labov, 1963).

Theoretical work in sociolinguistics has, furthermore, established distinct categories of social identity or characteristics that a variant can be associated with, and this turns out to have important consequences for whether model-based bias can be expected to operate. This work derives from the sociolinguistic framework of indexicality, developed by Eckert, 2008, Silverstein, 2003. Silverstein (2003) distinguished between different orders of indexicality that a cultural variant may be associated with. First-order indices identify a speaker as belonging to a particular group (defined, for example, by geographical area or socioeconomic class), but are not necessarily subject to social evaluation or even social awareness. For instance, Philadelphia English is characterized by a number of features, such as the pattern of pronunciation of the vowel in day vs. days, that remain unnoticed by the vast majority of people, including Philadelphians themselves. A shift to second-order indexicality occurs when variants come to index perceived traits of a group. As Eckert (2008, p. 436) put it:

A first-order index simply indexes membership in a population […] But the social evaluation of a population is always available to become associated with the index and to be internalized in speakers’ own dialectal variability to index specific elements of character. At that point, the linguistic form becomes a marker, a second-order index³

Indices combine with other features (both linguistic and non-linguistic) to create what Eckert termed an indexical field, “a constellation of meanings that are ideologically linked” (p. 464). Both first- and second-order indices are continuously open to a process of social reinterpretation, meaning that any index can give rise to a new higher-order index, whose importance relative to other indices shifts according to context (Eckert, 2008). As this suggests, different orders of indexicality should not necessarily be interpreted as being related to each other in a straightforwardly linear way.

Nevertheless, variants with higher-order indexicality can be considered to differ from first-order variants along two dimensions. The first dimension concerns the alienability of the trait indexed. While first-order variants index membership in a population (like “female” or “Midwesterner”), which is hard to acquire for non-members, higher-order variants index the alienable characteristics (like “sweet” or “naive”) that are perceived to belong to that population, but are available to non-members. That is to say, while Midwesterners might be stereotyped as sweet and naive, and New Yorkers as the opposite, it is entirely possible for a New Yorker to be naive or sweet, and for an Iowan to be neither. Alienable characteristics, in other words, are those that are not inherent to a population and may be possessed or acquired by members of any population. The second dimension concerns social relevance. The traits indexed by higher-order variants are not arbitrary characteristics of the group in question; rather, they are socially relevant characteristics set up in opposition to the perceived traits of other social groups. For instance, a first-order linguistic feature that indexes working-class men may also come to index the more alienable trait of toughness in populations where toughness is relevant to the distinction between working-class and middle-class men (cf. Connell, 1995, p. 55). In summary, we interpret the difference between higher-order indices and first-order indices in Eckert’s (2008) account in terms of two components: Higher-order indices are associated with traits that are both (a) alienable and (b) socially relevant, while first-order indices are associated with traits that are neither.

While Eckert (2008) emphasized the contextuality and fluidity of the distinction between first- and second-order indices, it does not necessarily follow that there is no clear boundary between them (indeed, the experimental results we present in this paper support a rather strong boundary), or that it is a theoretical division that lacks real-world consequences. One such consequence is that variants at different orders of indexicality can be expected to change differently. Those with first-order indexicality alone change without regard to social evaluation, and speakers may not even be aware that they exist. Those with second- or higher-order indexicality, by contrast, may become socially stigmatized (Labov, 1972) or may acquire prestige, including covert prestige in which they are used to index a trait (such as “violent”) that is stigmatized in mainstream culture (Trudgill, 1972). The social stigmatization or prestige is, in turn, likely to have an effect on the propagation of the variant, making the distinction between orders of indexicality important from a cultural evolutionary point of view. Variants with first-order indexicality are more likely to spread through neutral forces. Once they acquire higher-order indexicality, selection—due to model-based bias—is likely to start operating, and to the extent that higher-order traits are in some sense desirable, they become attractive candidates for adoption outside their original population.

Ethnographic case studies in sociolinguistics have demonstrated the importance of both alienability and social relevance to acquiring a second-order linguistic feature. (Sneller, 2014, Sneller, submitted for publication), for instance, found white working-class heterosexual men adopting TH-fronting (the pronunciation of the phoneme /θ/ as [f], as in [bof] for both) from their African-American neighbors. (Sneller, submitted for publication) argued that for these speakers, TH-fronting was no longer a first-order index of membership in the African-American community, and instead had come to index the second-order trait of street or toughness. Toughness is an alienable trait and, for these speakers whose identity depended on their readiness to fight, it was a socially relevant one. Importantly, neither the female interviewees nor the one gay male interviewee in the study exhibited any TH-fronting. For these speakers, toughness might have been alienable but was not socially relevant, making TH-fronting less accessible for adoption.

The theoretical framework of indexicality thus suggests a hypothesis that seems to be supported—but cannot be strictly tested—by ethnographic fieldwork such as that of Sneller, 2014, Sneller, submitted for publication. The hypothesis is as follows: Cultural variants propagate more readily when they have second-order indexicality (i.e., index traits that are both alienable and socially relevant) than when they have first-order indexicality (i.e., index traits that are neither alienable nor socially relevant). This hypothesis can be broken down into two sub-questions:

• Do variants with second-order indexicality propagate between populations more readily than variants with first-order indexicality?

• If the answer to Question 1 is yes, does the effect depend on both components of second-order indexicality being present, or do alienability and social relevance contribute independently?

In this paper we present two experiments designed to test our hypothesis. In Experiment 1 we manipulated alienability in a socially relevant trait and tested whether it spread more readily to a new population (Section 2). In Experiment 2 we replicated Experiment 1, but removed the social relevance of the trait (Section 3). Taken together these two experiments allow us to answer both the questions posed above. Table 2 summarizes conditions across the two experiments. We compare the two experiments in detail in Section 4.