Original ArticlesSocial network size can influence linguistic malleability and the propagation of linguistic change
Introduction
Imagine that you are trying to come up with a name for your band, and you are debating between Karaoke Dentist and Popcorn Logic.1 If you were to ask one of your friends which name they prefer and they responded Karaoke Dentist, this might tilt you towards choosing this name. In contrast, if you were to ask twenty-one of your friends, and ten of them were to prefer Popcorn Logic and eleven, including that friend, had preferred Karaoke Dentist, this friend’s preference of Karaoke Dentist is likely to not influence you that much. In other words, there is an inverse relationship between how many sources one has and how informative each source is. This relationship between sample size and informativity is a general principle and likely extends to linguistic information as well. Therefore, people who are exposed to linguistic input from many sources should be less susceptible to the influence of new incoming linguistic input compared with people who only interact with few people. Throughout this paper, the number of people someone regularly interacts with would be referred to as the person’s social network size, and similarly, people who interact with many people regularly would be referred to as people with large social networks. The hypothesis that this paper tests then is that the larger people’s social network, the less they would be influenced by exposure to a new speaker. Such an argument has implications not only for our understanding of how people learn and update their knowledge, but also for language change, as it suggests that the spread of linguistic change might depend more on people with smaller rather than larger social networks. Study 1 test whether people’s social network size influences the degree to which they are susceptible to the influence of a new speaker, and Study 2 describes simulations that test whether such differences in malleability could lead people with smaller social networks to be important for the propagation of linguistic change.
When people interact, their language tends to align across all linguistic levels (e.g., Giles, Coupland, & Coupland, 1991). For example, it has been found that during interaction people accommodate their pitch, speech rate, frequency and duration of pauses, standardness of speech, lexical choices, grammatical choices, and even nonverbal mannerisms, to those of their interlocutor (Branigan et al., 2000, Brennan and Clark, 1996, Chartrand and Bargh, 1999, Coupland, 1980, Gregory and Webster, 1996, Jaffe and Feldstein, 1970, Street, 1982, Thakerar et al., 1982). In fact, even though social factors seem to modulate some of these effects (e.g., Babel, 2012, Giles et al., 1991, Gregory and Webster, 1996), passive exposure without any interaction can also increases alignment (e.g., Bock, 1986, Goldinger, 1998). Such alignment has been theorized to reflect learning and thus lead to long-term convergence (Bock and Griffin, 2000, Chang et al., 2006). For example, the speech of previously-unfamiliar college roommates has been shown to become more similar after living together (Pardo, Gibbons, Suppes, & Krauss, 2012). More generally, it has been argued that we use incoming input to update our priors, and thus, as the statistics of our environment change, so do our representations (Jaeger & Snider, 2013).
Language learning can be seen as a type of social learning. One typical characteristic of social learning is the conformity bias, that is, preferentially copying behaviors that are frequent in the population (e.g., Boyd & Richerson, 2005). Importantly, people are not simply sensitive to the raw frequency of a certain behavior, but also the number of sources that exhibit it. Having five different friends vote for Karaoke Dentist is more informative than having the same friend vote for Karaoke Dentist five times. Indeed, even chimpanzees (though not orangutans) are more likely to adopt a behavior when it is performed by two out of three demonstrators, than when it is similarly performed two thirds of the time, but always by the same demonstrator (Haun, Rekers, & Tomasello, 2012). Three to 6 year-old children have also been found to imitate an action more when it is performed by two demonstrators than when it is performed by a single demonstrator twice (Herrmann, Legare, Harris, & Whitehouse, 2013).
One consequence of gathering information across many sources is that the weight ascribed to each source should decrease with the increase in number of sources. Consequently, the same input should be weighed differently by people who are exposed to different number of sources. Note that this argument doesn’t necessitate that sources are given equal weight. Even if we assume that we ascribe greater weight to the input that we receive from some people than from others, it is still the case that, on average, sources should be assigned lower weight the more of them we have. This should lead people with smaller social networks to assign greater weight to each person they encounter, and therefore to be more susceptible to each person’s influence.
Some initial evidence suggests that this is the case. In particular, one previous study found that the smaller the participants’ social network, the more they shifted their phonological boundary between /d/ and /t/ following exposure to a speaker with atypical productions (Lev-Ari, 2017). Furthermore, a control condition which tested participants on their learning of the phonological boundary of the exposure speaker and not on the change in their own boundary, ensured that participants’ ability and motivation to learn the speaker’s speech pattern did not depend on their social network size. That is, participants with large and small social networks learned the speaker’s speech patterns equally well, but those with larger social networks were less likely to generalize it and adjust their own general representation. This previous study thus suggests that social network size can influence how susceptible people are to the influence of new speakers. The current study goes beyond the previous findings in several ways. First, the previous study focused on perception. In order to link representational malleability to language change, it is important to examine the influence of social network size also on production. The current study tests the influence of social network size on both prediction and production. Additionally, the previous study focused on the phonological level, whereas this study focuses on the lexical level.
Languages constantly change. The word douchebag, the use of because to introduce a noun-phrase, a shift towards more constructions over -er constructions for comparatives, and speaking with a vocal fry are all instances of linguistic innovations that have gained popularity in recent years. In general, language is not a uniform phenomenon, but consists of great heterogeneity of variants and patterns. In many cases, however, this variation does not lead to linguistic change, as the variants do not propagate through the community (e.g., Weinreich, Labov, & Herzog, 1968). This paper proposes that people with smaller social network might play a particularly important role in propagating the diffusion of linguistic variants. This proposal might provide a partial solution to the threshold problem (Nettle, 1999) – the puzzle regarding how innovations, which are rare by definition, can spread through the community when speakers tend to use the most common variant they have encountered. One way of overcoming this problem is by assuming that speakers do not simply copy the most frequent variant, but that they hold some biases, leading them to be more likely to copy variants that are better in some way or that are used by more prestigious speakers (Nettle, 1999). The hypothesis tested in this paper adds that the threshold problem is also easier to overcome by speakers with small networks.
Previous research on linguistic diffusion tended to focus on identifying the innovators rather than those propagating the innovation (e.g., Fagyal et al., 2010, Labov, 2001, Milroy and Milroy, 1985). Interestingly, those who did investigate diffusion, especially diffusion of information and behavior, assigned a central role to diffusion via weak ties, that is, via relationships that are low in frequency and intensity (Bakshy et al., 2012, Granovetter, 1973, Mühlenbernd and Franke, 2012, Weimann, 1982). This research thus shows that non-central members could be crucial for the diffusion of behavior, and suggests that people with small social networks could play an important role in diffusing linguistic change despite their non-central role in the community.
Section snippets
Experiment 1
The aim of Experiment 1 is to test whether individuals with smaller social networks have more malleable linguistic representations. The malleability of individuals' linguistic representations was measured by testing the degree to which their general lexical boundary between some and many has changed as a consequence of exposure to a speaker whose lexical boundary differs from their own. Yildirim, Degen, Tanenhaus, and Jaeger (2016) have shown that people can learn a speaker’s boundary between
Experiment 2
Experiment 2 tests whether the findings from Experiment 1 could have implications for the process of language change. Specifically, Experiment 2 assumes the results of Experiment 1, namely, that people with smaller social networks have more malleable representations. It tests whether that could lead them to play an important role in the propagation of linguistic change despite the fact that they are, by definition, connected to fewer people, and therefore, could have a direct influence on fewer
General discussion
Together, these studies show that individuals with smaller social networks have more malleable representations, and that this greater malleability can lead them to play an important role in the propagation of linguistic change. Specifically, Experiment 1 shows that those with smaller social networks are more likely to adjust their lexical boundaries following exposure to new input. These results are in line with the proposal that the inverse relationship between number of sources and the
References (39)
Evidence for phonetic and social selectivity in spontaneous phonetic imitation
Journal of Phonetics
(2012)Syntactic persistence in language production
Cognitive Psychology
(1986)- et al.
Syntactic co-ordination in dialogue
Cognition
(2000) A novelty effect in phonetic drift of the native language
Journal of Phonetics
(2013)- et al.
Centers and peripheries: Network roles in language change
Lingua
(2010) - et al.
Majority-biased transmission in chimpanzees and human children, but not orangutans
Current Biology
(2012) - et al.
Stick to the script: The effect of witnessing multiple actors on children’s imitation
Cognition
(2013) - et al.
Alignment as a consequence of expectation adaptation: Syntactic priming is affected by the prime’s prediction error given both prior and recent experience
Cognition
(2013) Using Social Impact Theory to simulate language change
Lingua
(1999)- et al.
Phonetic convergence in college roommates
Journal of Phonetics
(2012)
Evaluation of noncontent speech accommodation
Language and Communication
Talker-specificity and adaptation in quantifier interpretation
Journal of Memory and Language
Emergence of scaling in random networks
Science
The persistence of structural priming: Transient activation or implicit learning?
Journal of Experimental Psychology: General
The origin and evolution of cultures
Conceptual pacts and lexical choice in conversation
Journal of Experimental Psychology: Learning, Memory, and Cognition
Becoming syntactic
Psychological Review
The chameleon effect: The perception-behavior link and social interaction
Journal of Personality and Social Psychology
Cited by (25)
An ERP investigation of accented isolated single word processing
2022, NeuropsychologiaCitation Excerpt :It is well documented that unfamiliar accents can lead to difficulties in speech comprehension (Munro and Derwing, 1995; Schmid and Yeni-Komshian, 1999; Anderson-Hsieh and Koehler, 1988; Lev-Ari, 2018; Major et al., 2002).
Syntactic Adaptation
2018, Psychology of Learning and Motivation - Advances in Research and TheoryCitation Excerpt :In addition, social context may be important for changes in the learner's own language production. If the learner identifies certain input as coming from an “in group”, or group they want to belong to, the set of probability distributions derived by adaptation to this group may form the learner's default, own-use model of language (for social effects on priming and adaptation, see e.g. Chun, Barrow, & Kaan, 2016a, 2016b; Lev-Ari, 2016, 2018; Weatherholtz, Campbell-Kibler, & Jaeger, 2014). Another difference observed between adaptation in the lab and language learning is that adaptation in the lab seems to be restricted to the specifics of the context, whereas a language learner is able to generalize over situations.
The semiotics of consensus: Impact of network topology on communication strategies in spanish language interaction networks
2023, The Semiotics of Consensus: Impact of Network Topology on Communication Strategies in Spanish Language Interaction NetworksScale in Language
2023, Cognitive Science