law search

The process of searching for relevant legal materials is fundamental to legal reasoning. However, despite its enormous practical and theoretical importance, law search has been given inadequate attention by scholars. A computational approach to studying law search generates a mathematical model of search and then tests how well that model approximate human-generated data.

bending the law

Bending the Law is an open source web platform that allows users to navigate a subset of Supreme Court cases (1950-2001) using a new representation of the legal corpus as a structured “multi-network” based on citation and semantic similarity (as represented by a topic model). Users enter the geometric space through any identified case, which itself may be generated through keyword searches on publicly available or proprietary databases. The user interface will introduce users to other cases in the “vicinity” of that pre-identified case, based on weights placed on types of relationships between cases (citations by, citations of, and topical similarity). The search is meant to return cases that will be strong candidates for precedential reasoning. This is accomplished via a new search-based algorithm that returns a ranked list of geometrically closest cases (using a search-based metric) as well as a map of the space, centered on a case of origin. Note that the cases are clickable and linked to the CourtListener Database. The “map” is generated by clicking on “View Case Network” and shows not only the relation of cases to the initial query, but also the relations of the closest cases to each other. This is a prototype for a larger effort.

View the search engine

Developers: Reed Harder (Dartmouth College); Greg Leibon (Dartmouth College); Michael Livermore (UVA); Allen Riddell (Dartmouth College), and Dan Rockmore (Dartmouth College)

We acknowledge the partial support of the Dartmouth College Department of Mathematics and Neukom Institute for Computational Science