CAREER: Discourse Processing and Content Generation for Document Simplification

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。简化是使文本更容易为目标受众（如语言学习者、儿童和有语言障碍的个人）所理解，同时保留其意义和内容的过程。缺乏可获取的材料会加剧社会问题，例如，大学录取和经济援助申请中使用的语言的复杂性导致新兴双语学生接受高等教育的机会滞后；鉴于医疗错误信息的增加，特别是在2019冠状病毒病大流行之后，世卫组织认识到获取技术信息的紧迫性。虽然在句子简化方面已经做了很多工作，但很少有数据集大到足以训练监督模型；简化文档还涉及不同于句子级别的操作，包括内容添加，以及句子如何相互连接。该项目旨在为文件简化开发新的资源和数据驱动的方法，有可能解决一系列高风险领域的信息透明度和公平访问问题。该项目还将支持教育和培训不同学科的本科生和研究生。为了实质性地推进文档简化，这个CAREER项目将处理现有简化工作中的几个关键问题，包括语料库多样性、解释生成和文档级方法。这是通过以下研究活动实现的：(1)引入新的语料库，解决简化研究中数据多样性的紧迫挑战，并实现新的应用场景，特别是在技术和术语文本的可访问性方面；(2)在简化过程中处理内容的添加和细化——这是一个以前很少探索的挑战，并提出一个新颖的、语言信息丰富的框架，以表征和产生细化；(3)开发基于话语结构的文档简化模型，同时使用连贯结构和实体显著性。整合话语的创新方法对模型提出了更大的挑战，即考虑到话语的延伸。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。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).Simplification is the process of making a text more accessible to a target audience, e.g., language learners, children, and individuals with language impairments, while preserving its meaning and content. The lack of accessible material can exacerbate social issues, for example, the complexity of language used in college admission and financial aid applications has contributed to the lagging access to higher education among emergent bilingual students; the WHO has recognized the urgency of accessible technical information, given the rise of medical misinformation especially in the wake of the COVID-19 pandemic. While there has been much work on sentence simplification, very few datasets are large enough to train supervised models; simplifying a document also involves different operations from those at the sentence level, including content addition, and how sentences connect with each other. This project aims to develop new resources and data-driven approaches for document simplification, with the potential to address information transparency and fair access across a range of high-stake domains. This project will also support the education and training of a diverse body of undergraduate and graduate students across disciplines.To substantially advance document simplification, this CAREER project will tackle several key issues in existing simplification work, including corpora diversity, explanation generation, and document-level approaches. This is achieved by the following research activities: (1) introducing new corpora that tackle the pressing challenge of data diversity in simplification research and enable new application scenarios, especially in the accessibility of technical and jargon-laden texts; (2) tackling content addition and elaboration during simplification---a previously little-explored challenge, and propose a novel, linguistically-informed framework that characterizes and generates elaborations; (3) develop models for document simplification that are informed by structures of discourse, using both coherence structure and entity salience. The innovative ways to integrate discourse target a larger challenge for models to take stretches of discourse into account.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.