Considerations for the social impact of natural language processing

Abstract

Natural language processing (NLP) technologies have transformed how people access information and communicate with one another. It has thus become critical to take stock of the social impact of natural language processing technologies. In this thesis, I review practices at different stages of development for NLP systems and examine some of the issues that arise in turn, considering the social and political contexts that shape how systems are developed and deployed. This thesis contributes three case studies of natural language processing technologies which exemplify many of the key issues in data collection practices and real-world system usage. The first two case studies situate computational models of text and machine translation in the complex social and political contexts that have informed the development of these applications. The third case study involves a reflection on original work in building and evaluating a system for representing biomedical relationships learned from text. In addition to the findings from these case studies, I contribute a practice-based framework for reflecting on factors that influence social impact at various stages of NLP system development.