Societies of strangers do not speak grammatically simpler languages (original submission)

Abstract

Many recent proposals claim that languages adapt to their environments. The Linguistic Niche hypothesis claims that languages with numerous native speakers and substantial proportions of non-native speakers (societies of strangers) will tend to lose grammatical distinctions. In contrast, languages in small, isolated communities should maintain or expand their range of grammatical markers. Here, we test such claims using a new global dataset of grammatical structures - Grambank. We model the impact of the number of native speakers, the proportion of non-native speakers, the number of linguistic neighbors, and the status of a language on grammatical complexity while controlling for spatial and phylogenetic autocorrelation. We deconstruct "grammatical complexity" into two separate dimensions: (i) how much morphology a language has ("fusion"), and (ii) the amount of information obligatorily encoded in the grammar ("informativity"). We find several instances of weak positive associations but no inverse correlations between grammatical complexity and sociodemographic factors. Our findings cast doubt on the widespread assumption that grammatical complexity is shaped by the sociolinguistic environment.

Sociolinguistic Typology Meets Historical Corpus Linguistics

“…(2023) and Shcherbakova et al. (2023). Typically, such studies are quantitative and correlational: they take data from databases of present‐day languages such as WALS (Dryer & Haspelmath 2013) or Grambank (Skirgård et al.…”

“…One way to scale up Trudgill's method is to use databases containing information about a large number of languages, and this has been done by, for example, Lupyan & Dale (2010), Bentz & Winter (2013), Sinnemäki & Di Garbo (2018, Koplenig (2019), Sinnemäki (2020), Kauhanen et al (2023) and Shcherbakova et al (2023). Typically, such studies are quantitative and correlational: they take data from databases of present-day languages such as WALS (Dryer & Haspelmath 2013) or Grambank (Skirgård et al 2023), operationalise sociolinguistic information (such as population size or proportion of L2 speakers), and assess the relationship between these variables statistically. Sinnemäki & Di Garbo (2018), for instance, look at verbal inflectional synthesis in a dataset of 309 languages, finding that number of L1 speakers has a significant effect on degree of synthesis, with a borderline significant effect of proportion of L2 speakers as well.…”

Languages with more speakers tend to be harder to (machine-)learn

“…Yet, depending on how you count, there are between 6,000 and 8,000 different languages and language varieties on the planet [13][14][15] that vary greatly in their structural properties. 16,17 A growing body of cross-linguistic research has begun to document that the natural and social environments in which languages are being used and learned drive this diversity [18][19][20][21] , that language structure is influenced by socio-demographic factors such as the estimated number of speakers 18,[21][22][23] and that that the long-held belief of a principle of "invariance of language complexity" 24 may be incorrect. 25 In this article, we investigate another long-held assumption that, to the best of our knowledge, has not been systematically tested yet: the assumption that all languages are equally difficult to learn.…”

“…25 . To rule out that the results are driven by these limitations, we create two fully balanced and parallel multilingual corpora, which we use to train seven different LMsranging from very simple n-gram models to state-of-the-art deep neural networksand measure how difficult it is for each LM to build an adequate representation of the input.Importantly, previous research [36][37][38][39][40][41]21 has shown that cross-linguistic (and cross-cultural) studies that seek to analyse potential statistical associations between language features and external factors must take into account Galton's problem, which refers to the potential confounding of…”

“…Importantly, previous research [36][37][38][39][40][41]21 has shown that cross-linguistic (and cross-cultural) studies that seek to analyse potential statistical associations between language features and external factors must take into account Galton's problem, which refers to the potential confounding of…”