A computational approach for modeling the indexical field
This talk aims to discuss how multiple social meanings come to be associated with a linguistic variable, in what Eckert (2008: 454) has called “an indexical field, or constellation of ideologically related meanings.” Since its proposal, the concept has been fruitfully applied in several works (e.g. Campbell-Kibler 2009, Becker 2013, Walker et al 2014, Tyler 2015) to explain ideological inter-relations among variables’ potential meanings. However, although representations of indexical fields are deeply rooted in empirical evidence, there isn’t yet an objective method for computing and reliably replicating them across studies.
I will base this discussion on the results of a matched-guise experiment (Lambert et al 1960, Campbell-Kibler 2009) aimed at uncovering the social perceptions of variable coda /r/ in São Paulo Portuguese, applied to 185 residents of the city. The results show that coda /r/ is strongly associated with geographical identities – namely, the tap with the capital city and retroflex /r/ with the countryside –, from which further inferences arise on speakers’ social status regarding their social class, area of residence, level of education, along with personal traits such as being “articulate” and “hardworking.” I will explore certain interactions between variable /r/ and participants’ social profiles, as well as the fact that certain plausible correlations don’t arise, in comparison to previous perception studies. Finally, I will present an objective and falsifiable method, Minimum Spanning Trees (Gower; Ross 1969), which allows for a visual representation of variable /r/’s indexical field based on the matched-guise data.
In addressing the problem of replicability and falsifiability in representing indexical fields, I will argue that a computational model allows not only for more objective interpretation of observed correlations, but also for predicting which new meanings are more likely to be attributed to sociolinguistic variables by different social groups, thus reinforcing the dynamic yet structured nature of sociolinguistic perceptions and evaluations.