Prototype Synthesis for Model Laws

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 1579–1588, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Prototype Synthesis for Model Laws Matthew Burgess University of Michigan mattburg@umich.edu Eugenia Giraudy UC Berkeley & YouGov eugenia.giraudy@gmail.com Eytan Adar University of Michigan eadar@umich.edu Abstract State legislatures often rely on existing text when drafting new bills. Resource and expertise constraints, which often drive this copying behavior, can be taken ad- vantage of by lobbyists and special inter- est groups. These groups provide model bills, which encode policy agendas, with the intent that the models become actual law. Unfortunately, model legislation is often opaque to the public–both in source and content. In this paper we present LOBBYBACK, a system that reverse en- gineers model legislation from observed text. LOBBYBACK identifies clusters of bills which have text reuse and gener- ates “prototypes” that represent a canon- ical version of the text shared between the documents. We demonstrate that LOBBY- BACK accurately reconstructs model leg- islation and apply it to a dataset of over 550k bills. 1 Introduction Beginning in 2005, a number of states began passing “Stand Your Ground” laws–legal protec- tions for the use of deadly force in self-defense. Within a few years, at least two dozen states im- plemented a version of the this legislation (Gar- rett and Jansa, 2015). Though each state passed its own variant, there is striking similarity in the text of the legislation. While seemingly “viral” the expedient adoption of these laws was not the result of an organic diffusion process, but rather more centralized efforts. An interest group, the American Legislative Exchange Council (ALEC), drafted model legislation (in this case modeled on Florida’s law) and lobbied to have the model law enacted in other states. While the influence of the lobbyists through model laws grows, the hidden nature of their original text (and source) creates a troubling opacity. Reconstructing such hidden text through analy- sis of observed, potentially highly mutated, copies poses an interesting and challenging NLP prob- lem. We refer to this as the Dark Corpora prob- lem. Since legislatures are not required to cite the source of the text that goes into a drafted bill, the bills that share text are unknown beforehand. The first problem therein lies in identifying clusters of bills with reused text. Once a cluster is identi- fied, a second challenge is the reconstruction of the original or prototype bill that corresponds to the observed text. The usual circumstances under which a model law is adopted by individual states involves “mutation.” This may be as simple as modifying parameters to the existing policy (e.g., changing the legal limit allowed of medical mar- ijuana possession to 3.0 ounces from 2.5) or can be more substantial, with significant additions or deletions of different conditions of a policy. Inter- estingly, the need to maintain the objectives of the law creates a pressure to retain a legally meaning- ful structure and precise language–thus changes need to satisfy the existing …

Content truncated to 500 words