People express themselves in different ways due to their background, their intended audience, the conventions of their language for the genre in question, or just as a matter of personal style. This variation has important consequences for low-level Natural Language Processing (NLP) tasks as well as downstream text classification applications. At the same time it directly enables the reverse tasks of predicting variables such as speaker demographics, native language, target audience, etc. Leveraging these stylistic differences in NLP can be beneficial for machines and humans alike, while on the other hand bringing the risk of misinterpretation and introduction of new biases, as I have been investigating in my work at the Positive Psychology Center at UPenn and the Ubiquitous Knowledge Processing Lab in Darmstadt.

Watch the webcast replay >>