Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier


Open Access Dissertation

Document Type


Degree Name

Doctor of Philosophy (PhD)

Degree Program

Computer Science

Year Degree Awarded


Month Degree Awarded


First Advisor

Brendan O'Connor

Subject Categories

Artificial Intelligence and Robotics | Data Science


The digital age has dramatically increased access to large-scale collections of digitized text documents. These corpora include, for example, digital traces from social media, decades of archived news reports, and transcripts of spoken interactions in political, legal, and economic spheres. For social scientists, this new widespread data availability has potential for improved quantitative analysis of relationships between language use and human thought, actions, and societal structure. However, the large-scale nature of these collections means that traditional manual approaches to analyzing content are extremely costly and do not scale. Furthermore, incorporating unstructured text data into quantitative analysis is difficult due to texts’ high-dimensional nature and linguistic complexity.

This thesis blends (a) the computational strengths of natural language processing (NLP) and machine learning to automate and scale-up quantitative text analysis with (b) two themes central to social scientific studies but often under-addressed in NLP: measurement—creating quantifiable summaries of empirical phenomena—and causal inference—estimating the effects of interventions. First, we address measuring class prevalence in document collections; we contribute a generative probabilistic modeling approach to prevalence estimation and show empirically that our model is more robust to shifts in class priors between training and inference. Second, we examine cross- document entity-event measurement; we contribute an empirical pipeline and a novel latent disjunction model to identify the names of civilians killed by police from our corpus of web-scraped news reports. Third, we gather and categorize applications that use text to reduce confounding from causal estimates and contribute a list of open problems as well as guidance about data processing and evaluation decisions in this area. Finally, we contribute a new causal research design to estimate the natural indirect and direct effects of social group signals (e.g. race or gender) on conversational outcomes with separate aspects of language as causal mediators; this chapter is motivated by a theoretical case study of U.S. Supreme Court oral arguments and the effect of an advocate’s gender on interruptions from justices. We conclude by discussing the relationship between measurement and causal inference with text and future work at this intersection.


Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.