By

Large-scale dialect surveys have long been a fundamental component of sociolin-guistics and variation studies, but they have traditionally required significant investments of time and resources to collect relatively small amounts of data. In this study, I exam-ine whether textual corpora collected from the Internet, particularly the social-networking website Twitter, can be used to conduct such surveys more quickly with less effort. I dis-cuss the utility of Twitter as a linguistic data source, explain the computational and linguis-tic methods necessary to collect and process worthwhile data, and use corpora from Twit-ter to plot the distribution of three regional variables in American English: soft drink ter-minology, the use of ‘hella ’ as an intensifier, and the morphosyntactic construction ‘needs X-ed’. I find that these corpora can replicate the findings of previous surveys (as seen with soft drink terms), reveal the existence of previously undocumented regional divides (as seen with ‘hella’), and highlight possible examples of ongoing linguistic diffusion (as seen with ‘needs X-ed’). I conclude that, although this method has issues with tracking certain types of sociolinguistic variables, it is still a promising source of data for conducting rapid studies in many areas of regional linguistic variation. ii Acknowledgements This study is the result of innumerable contributions. I wish to thank first and foremost my advisor, Kathryn Campbell-Kibler, and my advising committee of Brian Joseph and Chris Brew. The OSU Changelings and Clippers discussion groups, the attendees of GURT 2011