Variation-Based Distance and Similarity Modeling: A Case Study in World Englishes

Abstract

Inspired by work in comparative sociolinguistics and quantitative dialectometry, we sketch a corpus-based method (Variation-Based Distance & Similarity Modeling-VADIS for short) to rigorously quantify the similarity between varieties and dialects as a function of the correspondence of the ways in which language users choose between different ways of saying the same thing. To showcase the potential of the method, we present a case study that investigates three syntactic alternations in some nine international varieties of English. Key findings include that (a) probabilistic grammars are remarkably similar and stable across the varieties under study; (b) in many cases we see a cluster of "native" (a.k.a. Inner Circle) varieties, such as British English, whereas "non-native" (a.k.a. Outer Circle) varieties, such as Indian English, are a more heterogeneous group; and (c) coherence across alternations is less than perfect.

Using VADIS to weigh competing epicentral influence

“…In this paper, I apply VADIS to the study of AmE vs. BrE epicentral influence on CanE using the R package VADIS, downloadable from GitHub. Despite not having been conceived as a method to study linguistic epicentres, VADIS relates well to this field of study, seeing as it performs a multivariate analysis that measures ‘inter‐speaker variation by assessing the structure of intra‐speaker variability’ (Szmrecsanyi et al., 2019, p. 1). In doing so, it helps provide an answer to the following questions, or lines of evidence: (i) What are the (intra‐ and extra‐linguistic) factors that are statistically significant in determining a linguistic phenomenon?…”

“…This paper investigates competing epicentre influence between hyper‐central AmE and super‐central BrE (Mair, 2006) on the CanE target variety. To do so, it relies on a newly developed methodology by Szmrecsanyi, Grafmiller, and Rosseel (2019) that models probabilistic distance and similarity among language varieties and linguistic phenomena. While the first section outlined the research objectives and the subject of the present analysis, section 2 will present the data and methods used for data retrieval, annotation, and analysis.…”

Representations of Language Varieties Are Reliable Given Corpus Similarity Measures

“…A third computational approach leverages some model of language variation to measure relationships between different partitions of linguistic data. For example, recent work has measured the linguistic similarity between varieties of English, like New Zealand English vs. Australian English (Dunn, 2019a;Szmrecsanyi et al, 2019). When expanded across aligned corpora, these models can be used to determine if there is a consistent pattern of variation: does New Zealand English have the same distinctive lexical choices on the web that it has in tweets?…”

Axel Bohmann, Variation in English worldwide: Registers and global varieties (Studies in English Language). Cambridge: Cambridge University Press, 2020. Pp. xvii + 250. ISBN 9781108751339.

“…Most studies observe a dichotomy between L1 and L2 varieties (or Inner vs Outer Circle) (e.g. Szmrecsanyi et al 2019); sometimes we can also observe a North American cluster or detect the influence of British English on its former colonies (as in the case of Bohmann's study with British and New Zealand English and the North American varieties clustering together). Other variety-groupings are harder to interpret and explain on sociohistorical grounds.…”

“…Finally, Bohmann's study sets the pace for future systematic quantitative research in World Englishes that aims to compare varieties on linguistic grounds sampling from naturalistic language data (for a similar attempt but focusing on probabilities rather than frequencies and on a small set of linguistic variables see Heller 2018;Röthlisberger 2018;Szmrecsanyi et al 2019). What Bohmann's work and other quantitative research in World Englishes have in common is that the theoretical models of World Englishes that they draw on often fail as a perfect explanans for their findings.…”