CAREER: Large-Scale Exploration and Interpretation of Consumer-Oriented Legal Documents

人们在日常生活中面临着大量的文本，他们必须选择读什么，不读什么。为了在我们的信息社会中发挥作用，消费者必须阅读并接受服务协议条款、金融协议、医疗协议、租赁协议、隐私政策和其他各种各样的“细则”，才能获得必要的商品和服务。这些面向消费者的法律文件（简称cold）规定了需求、处罚、可接受使用的边界、出现问题时的追索权选项、隐私实践、知识产权规定以及许多其他重要主题。人们往往在没有阅读或理解的情况下接受感冒，而缺乏理解会削弱个人的权力，并对他们产生不平等的影响。该项目将回答三个相关问题：(1)感冒中存在哪些重复出现的信息结构和知识类型？(2)文本挖掘在cold信息自动提取中的能力和局限性？(3)感冒的内容在多大程度上与消费者的兴趣和需求相交叉？为了回答这些问题，该项目将开发自然语言处理方法，支持用户参与通常不引人入胜但重要的文本。此外，该项目将引入一年级本科生进行研究，并鼓励他们从事STEM职业。该项目将通过关注以下目标来推进知识。首先，该项目将通过从在线资源中创建大规模的感冒语料库来发现常见感冒类型的可用性和特征。这些语料库将使研究可用性、可访问性、导航和可读性等问题成为可能。其次，该项目将开发自动提取COLD文本的两个特别突出的特征的方法：选择点（文本中描述读者可能为其利益采取的行动的语句）和离群语句（偏离相关类型COLD的典型语句，将COLD与同类区分开来并激发强烈关注）。第三，该项目将构建浏览器扩展，以探索选择点和离群值陈述如何影响人们对感冒内容的参与。为了支持进一步的研究，该项目还将为自然语言处理和公共政策方面的研究人员制作和传播一系列语料库、语言模型和其他工具。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

People face large volumes of text in their everyday lives, and they must choose what to read and what to leave unread. To function in our information society, consumers must read and accept terms of service agreements, financial agreements, health care agreements, rental agreements, privacy policies, and other varieties of "fine print" to receive essential goods and services. These consumer-oriented legal documents (COLDs, for brevity) specify requirements, penalties, boundaries of acceptable use, options for recourse if something goes wrong, privacy practices, intellectual property stipulations, and many other important topics. People tend to accept COLDs without reading or understanding them, and the lack of understanding disempowers individuals and affects them unequally. This project will answer three related questions: (1) What recurring information structures and types of knowledge exist in COLDs?; (2) What are the capabilities and limitations of text mining applied to automating extraction of information from COLDs?; and (3) To what extent do the contents of COLDs intersect with the interests and needs of consumers? Toward answering these questions, the project will develop natural language processing methods that support user engagement with typically unengaging but important text. Additionally, this project will introduce first-year undergraduates to research and encourage them to pursue STEM careers.The project will advance knowledge by focusing on the following goals. First, the project will discover the availability and characteristics of common types of COLDs by creating large-scale corpora of them from online sources. These corpora will enable studying issues in availability, accessibility, navigation, and readability. Second, the project will develop methods for automated extraction of two particularly salient features of COLD text: choice points (statements in text that describe actions a reader can take potentially for their benefit) and outlier statements (statements that deviate from what is typical for the relevant type of COLD, differentiating a COLD from its peers and motivating acute attention). Third, the project will build browser extensions to explore how choice points and outlier statements affect people's engagement with COLDs' contents. To support further research, the project also will produce and disseminate an array of corpora, language models, and other tools for researchers in natural language processing and public policy.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.