Abstract:

Across its many forms, user-generated content (UGC) acts as a sample of all human discourse. It has been harder to automatically handle this type of text with traditional tools; unlike newswire, it comprises a much broader and more direct example of language. This talk examines the kind of variation we see in user-generated content, especially social media. Further, this talk not only presents methods for coping with the noise without changing it, but also goes on to explain the many kinds of latent information expressed by the stable, consistent linguistic variation seen across society and the internet.