First steps towards the detection of contact layers in Bangime: a multi-disciplinary, computer-assisted approach

Abstract

Bangime is a language isolate, which has not been proven to be genealogically related to any other language family, spoken in Central-Eastern Mali. Its speakers, the Bangande, claim affiliation with the Dogon languages and speakers that surround them throughout a cliff range known as the Bandiagara Escarpment. However, recent genetic research has shown that the Bangande are genetically distant from the Dogon and other groups. Furthermore, the Bangande people represent a genetic isolate. Despite the geographic isolation of the Bangande people, evidence of language contact is apparent in the Bangime language. We find a plethora of shared vocabulary with neighboring Atlantic, Dogon, Mande, and Songhai language groups. To address the problem of when and whence this vocabulary emerged in the language, we use a computer-assisted, multidisciplinary approach to investigate layers of contact and inheritance in Bangime. We start from an automated comparison of lexical data from languages belonging to different language families in order to obtain a first account on potential loanword candidates in our sample. In a second step, we use specific interfaces to refine and correct the computational findings. The revised sample is then investigated quantitatively and qualitatively by focusing on vocabularies shared exclusively between specific languages. We couch our results within archeological and historical research from Central-Eastern Mali more generally and propose a scenario in which the Bangande formed part of the expansive Mali Empire that encompassed most of West Africa from the 13th to the 16th centuries. We consider our methods to represent a novel approach to the investigation of a language and population isolate from multiple perspectives using innovative computer-assisted technologies.

Detecting Lexical Borrowings from Dominant Languages in Multilingual Wordlists

“…The second method, called Cognate-Based borrowing detection in the following, follows the approach by Hantgan et al (2022): it first computes cognates using a cluster-based approach for automated cognate detection in which words expressing the same concept whose average phonetic distance is below a certain threshold are assigned to the same cognate set (List et al, 2017), and then identifies all words assigned to cognate sets involving the dominant language as borrowings. We tested again normalized edit and SCA distances.…”

“…Zhang et al (2021) compare edit distance performance against SCA (List, 2012) distance performance, finding that SCA outperforms edit distance in accuracy. Hantgan et al (2022) build on this work, using dedicated methods for automated cognate detection applied to languages from different language families in order to identify clusters of related words resulting from lexical borrowing. List and Forkel (2022a) expand this work further, by applying a two-stage workflow in which they first identify language-family-internal cognates, using a method specifically apt for the detection of deep cognates, and then compute SCA distances between cognate sets from genetically unrelated languages in order to infer sets of words related by lexical transfer.…”

Bangime: secret language, language isolate, or language island? A computer‐assisted case study

“…As a result, we decided to withdraw the original article and instead share an updated and slightly shortened version of the study again for open peer review with Papers in Historical Phonology. We note that our approaches shared here have in part been superseded by novel techniques by now, notably Hantgan et al (2022) and List & Forkel (2022a). We hope, however, that the techniques described in this study are still of interest for scholars working on language isolates or languages heavily in luenced from language contact.…”

Open Problems in Computational Historical Linguistics

“…Quite a few new methods have been proposed to address the detection of lexical borrowings (Problem 2). Among these are supervised approaches (which I had deliberately excluded, since I consider unsupervised approaches as more useful when it comes to inference problems) that made use of recurrent neural networks ( Miller et al , 2020 ), there are tree-based approaches that even try to identify the direction of borrowings ( Neureiter et al , 2022 ), 6 and there are numerous attempts to handle very specific cases of lexical borrowing, such as contact across language families ( Hantgan et al , 2022 ; List & Forkel, 2022 ) or contact induced by dominant languages ( Kaiping & Klamer, 2022 ; Miller & List, 2023 ).…”