Natural Language Processing Support for eRulemaking

每年联邦监管机构发布4,000多条新规定。其中许多必须通过一个称为通知和评论（&NC）规则制定的过程来创建：行政机关起草一份拟议的规则，然后将该建议、任何基础数据及其法律的和政策依据公开征求公众意见。N C规则制定是当代公共政策制定中最重要的方法之一，也是最慢和最昂贵的方法之一。尽管行政机关可能会收到数十万条关于一项拟议规则的意见，但它的法律的义务是审查所有重要意见并作出答复。随着咨询、研究和/或证明要求的激增，规则制定者发现越来越难以跟踪这些要求，也越来越难以认识到哪些要求（如果有的话）与特定规则制定相关。电子规则制定（eRulemaking）有可能从根本上改变NC程序。它可以使这一过程更加透明，公众更容易了解，对机构来说，这一过程在实质上更加可靠，成本效益也更高。然而，到目前为止，电子备审系统和电子规则制定工作台只对现有技术进行了初步的利用。该基金将利用自然语言处理（NLP）的成熟和新兴方法来开发工具，以帮助机构规则制定者：（1）组织、分析和管理与拟议规则相关的评论、研究和其他支持文件;以及（2）分析拟议的规则，从大量法规和行政命令中标记可能相关的法律的授权，这些法规和行政命令可能需要在规则制定期间进行分析、磋商或认证。研究小组将与联邦运输部和商业部合作。 该团队将特别关注在监督和弱监督机器学习框架中使用信息提取，文本分类和面向意见的文本分析技术。评估将涉及：使用公认的NLP性能技术指标（例如，查全率和查准率）;定性和定量社会科学方法相结合，以评估机关各级工作人员对规则制定过程中工具整合的看法;以及受过法律培训、对规则制定过程有专业理解的研究人员的观察。这项研究将有助于实现电子规则制定的积极潜力，推进NLP的最新发展，并提高我们对技术对规则制定的影响的理解。由于其跨学科的组成-结合了自然语言处理的专业知识，有关监管法律和法律的信息系统的专家知识，以及技术对组织影响的社会科学经验-康奈尔大学的团队很好地为关键问题提供定性和定量数据，但在很大程度上还没有得到充分的研究，该项目为康奈尔大学信息科学专业的博士、硕士和本科生提供了一个重要的跨学科教育和研究机会。所有数据集和工具都将提供给其他研究人员。要开发的NLP方法是通用技术，可用于任何领域或流派，并且在需要管理，组织和分析大量文本的任何上下文中都很有用。最后，许多帮助机关规则制定者的技术也可用于设计机关网站，帮助公众在规则制定过程中搜索、分类和以其他方式有选择地获取材料。

Each year Federal regulatory agencies issue more than 4,000 new rules. Many of these must be created through a process known as notice and comment (N&C) rulemaking: the agency drafts a proposed rule and then exposes the proposal, any underlying data, and its legal and policy rationale to public comment. N&C rulemaking is one of the most important methods of contemporary public policy making; it is also one of the slowest and most expensive. Although an agency may receive hundreds of thousands of comments for a proposed rule, its legal obligation is to review and respond to all significant comments. As requirements to consult, study, and/or certify have proliferated, rule writers have found it increasingly difficult to keep track of them and to recognize which, if any, are relevant in a particular rulemaking. Electronic rulemaking (eRulemaking) has the potential to radically transform the N&C process. It could make the process more transparent and accessible to the public, and more substantively reliable and cost-effective for the agency. So far, though, E-docket systems and eRulemaking workbenches make only rudimentary use of available technology.This grant will use well-developed and emerging methods of natural language processing (NLP) to develop tools to aid agency rule writers in: (1) organizing, analyzing, and managing the comments, studies, and other supporting documents associated with a proposed rule; and (2) analyzing proposed rules to flag possibly relevant legal mandates from among the large number of statutes and Executive Orders that potentially requireanalyses, consultations, or certifications during rulemaking. The research team will collaborate with the Federal Departments of Transportation and Commerce. The team will focus, in particular, on the use ofinformation extraction, text categorization, and opinion-oriented text analysis techniques in both supervised and weakly supervised machine learning frameworks. Evaluation will involve: the use of accepted technical measures of NLP performance (e.g., recall and precision); a combination of qualitative and quantitative social science methods to assess integration of the tools into the rulewriting process as perceived by staff at various levels of the agency hierarchy; and observation by legally-trained researchers with expert understanding of the rulemaking process.Intellectual Merit. The research will help realize the positive potential of eRulemaking, advance the state-of-the-art in NLP, and improve our understanding of the effects of technology on rulemaking. Because of its interdisciplinary composition - combining expertise in NLP, expert knowledge about regulatory law and legal information systems, and social science experience in the effect of technology on organizations - the Cornell team is well situated to generate both qualitative and quantitative data about the crucial, but stilllargely under-studied, rulemaking process.Broader Impacts.The project provides an important opportunity for interdisciplinary education and research for PhD, master's, and undergraduate students in Cornell's Information Science Program. All data sets and tools will be made available to other researchers. The NLP methods to be developed are general-purpose techniques, trainable for any domain or genre, and useful in any context that requires managing, organizing, and analyzing large volumes of text. Finally, many of the same techniques that help agency rule writers can be used to designagency websites that help the public search, sort, and otherwise selectively access materials in the rulemaking process.