The Linguistics of Spin

There has been a great deal of recent research on the computational analysis of texts in order to identify sentiment; that is, to determine automatically whether a text expresses a positive or negative perspective or opinion on a topic. Most such work focuses on explicitly expressed opinions, e.g. automatically labeling movie reviews as favorable or unfavorable. In this talk, we will focus on a less studied problem: the computational analysis of implicit sentiment, or spin, in text. We identify underlying semantic properties of statements that predict perceptions of sentiment, and we show how observable syntactic reflexes of those properties can be used, fully automatically, to accurately label a text as expressing a positive or negative attitude toward its topic, even in the absence of overtly opinionated language.