@incollection{MoshnikovRykova2025, author = {Moshnikov, Ilia and Rykova, Eugenia}, title = {Collecting minority language data from Twitter (X): A case study of Karelian}, series = {Exploring digitally-mediated communication with corpora : Methods, analyses, and corpus construction}, booktitle = {Exploring digitally-mediated communication with corpora : Methods, analyses, and corpus construction}, editor = {Cotgrove, Louis and Herzberg, Laura and L{\"u}ngen, Harald}, publisher = {De Gruyter Brill}, isbn = {978-3-11-143401-8}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:526-opus4-20583}, publisher = {Technische Hochschule Wildau}, pages = {163 -- 186}, year = {2025}, abstract = {The visibility of an endangered language online plays a crucial role in language revitalisation. The internet offers a new domain for using minority languages, especially for speakers living outside the language communities. This article investigates Karelian language visibility on X, formerly known as Twitter, and describes the first corresponding data collection using language-related keywords and hashtags. In total, 2,625 entries written fully or partially in Livvi, South and Viena Karelian were scraped with Postman API. The visibility of Karelian on Twitter (X) has been increasing considerably in the past few years, with Livvi-Karelian being the most prominent dialect. Automatic language detection was tested on such data for Karelian for the first time, and allows the identification of Livvi-Karelian (or a mix of dialects that include Livvi-Karelian) with 99.7\% sensitivity, and South Karelian and Viena Karelian as Livvi-Karelian with 90\% and 73.8\% sensitivity, respectively. The entries were also analysed thematically, and 10 major topics were identified. Since the data was collected using keywords and hashtags related to the Karelian language itself, most of the entries are related to the language and vocabulary in sense of translation or language learning. Language status and policy is another important topic identified in the data. Although language-related topics are the most popular, there are a substantial number of entries on eight further topics. Excluding citations from religious texts and media headlines, 751 Twitter (X) entries could be used for linguistic and sociological research. Further data collection considerations are also discussed.}, language = {en} }