Beyond the Individual: Modeling Interpersonal Relationships in NLP Studies of Communication

Abstract: NLP studies of communication often focus on the individual: What we say, when we say it, and how we say it. Work on user prediction has shown that aspects of the individual (e.g., gender, age) are revealed in our language. Yet, the larger social context beyond the individual also plays an important role in our communication. In this talk, I will describe three studies from my group that model interpersonal relationships as a part of the social context in order to understand communication practices. In the first part of the talk, I will introduce a large-scale study of over 9M self-declared relationships in social media and show how relationships influence the what and when in our communications. Further, I will show how we can use these observed communications in social media to infer the likely type of relationship between individuals and, in turn, how knowledge of relationships is useful for predicting behavior like content resharing. In the second part, I will discuss recent work showing how we can directly incorporate social relationships in NLP models to change a message’s interpretation based on the social context in which it is communicated. Last, in the third part, I will discuss ongoing work at modeling empathy in interpersonal communication using appraisal theory. In this approach, we measure empathy by explicitly modeling the listener’s theory of mind based on how speaker and listener align in their communication.

The speaker: David is an assistant professor in the School of Information at the University of Michigan. He previously obtained his PhD at the University of California and was a postdoctoral researcher in the Department of Computer Science at Stanford University. His research is known for its interdisciplinary character and combines NLP with network analysis, computational social science, data science and psychology. This way, he tries to gain a 'social understanding' of both the content and the people involved in communication. 

Hosts: Sofie Labat  & Gabriella Lapesa. To arrange a meeting with a speaker and/or participate in the dinner after the talk, contact gabriella.lapesa@ims.uni-stuttgart.de

Colloquium for Computational Linguistics and Linguistics in Stuttgart