NLP

Narratives in Law and Politics: A Computational Linguistics Approach

Abstract

The goal of this research is to leverage data science tools to the challenge of measuring narratives in legal and political texts, and then to analyze the causes and consequences of those narratives in the political economy. This proposal outlines a system for taking the plain text of documents, extracting the latent narrative structures in the text, and extracting the narratives that are causally related to some cause or outcome. The system will be applied to an analysis of prejudicial narratives in law, politics, and mass media. A data-driven approach that can extract text-based measures of narratives would mark a major step forward in this field. Many phenomena of interest to historians and social scientists are embedded in text, so the development and application of natural language processing methods to the analysis of narratives will be useful to researchers in a variety of fields and settings. In more specific reference to SDSC goals, this research seeks to make progress on an interdisciplinary data science problem (measuring narratives and analyzing associated causal links) and then focus on applications in a social-science domain (political economy of law and mass media). The deep learning models described below require significant computational infrastructure and expertise. To accelerate the empirical analysis by domain social scientists, we seek support from SDSC in data storage and management, implementation of statistical and machine learning models, and the development of compelling data visualizations. Our team share a core belief in reproducibility and open science initiatives.

Starting Date

April 2020

PI / Partners

Center for Law and Economics (ETHZ)

Description

Problem:

  • This project aims at leveraging data science approaches to (i) extract narratives in legal and political texts, and (ii) analyze the causes and consequences of those narratives in political economy. Such objectives require application and adoptation of natural language processing tools as well as data-driven techniques for causal discovery and inference in high dimensional data.

Impact:

  • The findings – methods as well as estimates of the real-world impacts of narratives in the political economy – will be useful to social scientists and policymakers interested in the political and cultural factors underlying social disadvantage. Additionaly, working with text data introduces a challenge, leading to new research question to be tackled with inovative machine learning tools.

Bibliography

  • Angrist, J. and Pischke, J.-S. (2009). Mostly Harmless Econometrics: An empiricist’s companion. Princeton University Press, Princeton, NJ.

  • Ash, E., Chen, D. L., and Naidu, S. (2017a). Ideas have consequences: The impact of law and economics on American justice. Technical report, NBER.

  • Egami, N., Fong, C. J., Grimmer, J., Roberts, M. E., and Stewart, B. M. (2017). How to make causal inferences using texts.

  • Galletta, S., Ash, E., and Chen, D. L. (2019). Do judicial sentiments affect social attitudes?

  • Shiller, R. J. (2017). Narrative economics. American Economic Review, 107(4):967–1004.