NLP+CSS: Workshop on Natural Language Processing and Computational Social Science
Second workshop on NLP+CSS at ACL 2017
August 3 or 4, 2017, Vancouver, Canada
Includes an option for non-archival paper submissions, as well as invited talks, panels, and a poster session.
Organizers: David Bamman (UC Berkeley) A. Seza Doğruöz (Tilburg University), Dirk Hovy (U. of Copenhagen), David Jurgens (Stanford), Brendan O'Connor (UMass Amherst), Oren Tsur (Harvard/Northeastern), Svitlana Volkova (PNNL)
On Twitter: NLP and CSS @nlpandcss
Language is a crucial outcome from, and contributor to, complex social processes. What we say and how we say it has a mutual dependence with our social background and interactions, including factors like age, ethnicity, relationships, ideology, affect, etc. While the interdependence between language and social factors has received attention in both natural language processing (NLP) and the (computational) social sciences (CSS), these two fields still exist largely in parallel, holding back research insights and potential applications. This proposal aims to advance the joint computational analysis of social sciences and language by explicitly involving social scientists, NLP researchers, and industry partners.
Computational social science is a growing field that often seeks to develop insights from textual data, including not only novel corpora from online social interactions, but also historical records, news archives, speech transcripts, and many other sources. This young field is still in flux with respect to standard methodologies and theories. Although NLP is one of the main tools to analyze social behavior from text, the potential of natural language processing technologies and insights has still not yet been realized in CSS.
We face a similar challenge with integrating social factors into NLP research. Although characteristics of the communication partner, medium of communication (e.g. face-to-face or online), and context (e.g. presidential address vs. family conversation) influence language use, most NLP models do not account for any of these factors. This mismatch between modeling and reality has two effects: worse model performance on certain demographics or social groups, and a general degradation of model performance over time (data drift).
This pair of workshops seeks to fill these gaps by bringing together both NLP and social science researchers.
We invite research on any of the following general topics:
While all submissions will be reviewed equally, authors can optionally choose a non-archival submission, or else the standard archival submission option.