Understanding multilingualism in Switzerland using text mining algorithms

Abstract

Many of today’s societies are made up of multiple language groups, including groups of monolingual speakers and multilingual speakers of several different languages. We can ask many interesting questions about those societies including how widely each language is used, what topics are communicated in each language, whether there are time differences in the way information gets to each language group, and whether and how members of a language group communicate with members of another language group. We tackle these questions by looking at Switzerland, a highly multilingual society, with a large corpus of geotagged Twitter data. Specifically, we crawled 47 million tweets from 97,577 users, identified the language for each of those tweets, and analyzed those tweets using topic and language analysis tools. By using hierarchical Dirichlet scaling process, a nonparametric topic model for labeled data, we discover which topics are most popular for English, German, French monolinguals, as well as English-German, English-French, and German-French bilingual users. We analyze hashtags for major world events to understand whether certain groups have earlier access to information. We look at the general language use to compare the language variety of monolingual and bilingual users. By applying these computational methods to a large corpus of tweets from Switzerland, we show that there are many interesting linguistic and sociolinguistic phenomena that can be uncovered.