A corpus of regional American language from YouTube

Abstract

Recent years have seen an increase in the number of corpora of regional language variation for English, allowing new types of aggregate analysis to be conducted. While the creation of a corpus from written language material is relatively straightforward, transcribing speech is time-consuming, and thus there are no large corpora of transcribed American speech with broad geographic coverage. This paper describes the creation of a new corpus of regional American English from the automatically generated captions of videos from YouTube channels with a local American focus — mainly channels of regional and local government entities or civic organizations. The corpus, which consists of transcripts of over 29,267 hours of spoken language, will enable the analysis of regional patterns of lexical, morphosyntactic, and other types of variation in spoken American English. Exploratory analysis and mapping of the corpus data indicates regional variation in spoken language is evident.