ABSTRACT

With more than a quarter of the world's population communicating actively on social media (e.g. Facebook, Instagram, and Twitter), the field of Natural Language Processing has quickly jumped into tackling problems that are social, psychological, or medical in nature. Language on Facebook and Twitter has been used to measure human characteristics from individual depression and personality to community well-being and mortality rates. However, language analysis techniques were mostly developed for modeling documents or words rather than characteristics of people behind the language. Modeling introduces complexities such as handling heterogenous extralinguistic variables (demographics, socioeconomics), taking temporal information into account (forecasting future outcomes), and data are inherently multi-level (people, who belong to communities, generate documents).  On the other hand, traditional social scientific studies, often relying on surveys and lab experiments, are relatively restricted in scope and not grounded in ecological reality -- where computational studies can shine. This talk will cover 

(1) how language analysis techniques are being furthered to handle the characteristics of human-level analysis tasks, and 

(2) the social scientific technology enabled by language analyses can potentially improve and even save lives.