Digging into Human Rights Violations: Anaphora Resolution and Emergent Witnesses

“挖掘侵犯人权行为”项目将开发各种方法和工具，以发现和可视化收集的侵犯人权行为报告和证人证词中隐藏的受害者和身份不明的肇事者的故事。这种自然语言处理的研究是必要的，因为，a）记录侵犯人权行为的集合通常有数万到数千万个文件，这一规模掩盖了受害者，侵犯和侵犯者; B）许多受害者？s的故事只作为片段出现在观察员和幸存者的报告和叙述中; c）目前的工具擅长搜索，而不是阅读。这个项目的计算阅读器，使用为各种真相与和解和历史澄清委员会汇编的集合开发，将包括一个从分散在集合中的片段重建故事的引擎，以及一个用于导航这些汇编故事和它们来自的文件的界面。虽然开发的调查人员和检察官使用的侵犯人权的行为，这种新颖的工具将扩大文本挖掘的可能性，并将普遍有用的学者，研究人员和从业人员需要跟踪实体，事件和模式在大型文件集合。关键的发展是进一步研究回指解析（其中回指是指另一个表达式的存在-通常是其先行词的代词）。在某些情况下，识别回指相对简单，但在许多情况下，它提出了一个重大的挑战，传统的自然语言处理方法。该项目将涉及自然语言处理和定性语言分析相结合，其更广泛的影响在于，它显然有可能为从事人权问题以及真相与和解工作的学者和政府官员提供信息。此外，该项目还包括重要的学生培训，包括跨学科研究方法的培训。这笔赠款是作为Digging Into Data Challenge的一部分提供的，Digging Into Data Challenge是一项国际竞赛，旨在促进各国之间的研究合作，并鼓励采用创新方法分析社会科学和人文科学中的大型数据集。美国的研究人员将与加拿大的学者合作，以实现该项目的目标。

Digging into Human Rights Violations will develop methods and tools for discovering and visualizing the stories of hidden victims and unidentified perpetrators in collections of human rights violations reports and witness statements. This research in natural language processing is necessary because, a) collections documenting human rights violations frequently number many tens of thousands to tens of millions of documents, a scale that conceals victim, violation, and violator; b) many victim?s stories are only present as fragments in the reports and accounts of observers and survivors; and c) current tools excel at searching, not reading. This project's computational reader, developed using collections compiled for various Truth and Reconciliation and Historical Clarification Committees, will consist of an engine for reconstructing stories from fragments scattered across a collection, and an interface for navigating those compiled stories and the documents from which they come. Although developed for use by investigators and prosecutors of human rights violations, this novel tool will expand the possibilities of text mining and will be generally useful for scholars, researchers, and practitioners with a need to track entities, events, and patterns across large document collections. The key development is to further work on anaphora resolution (where anaphora is the existence of an expression referring to another -- usually a pronoun to its antecedent). In some cases identifying anaphora relatively straightforward, but in many cases it presents a significant challenge to traditional natural language processing methods. This project will involve a combination of natural language processing and qualitative language analysis.This project's broader impacts lie in its clear potential to inform scholars and government officials working on issues of human rights and in truth and reconciliation contexts. Further, the project includes significant student training, including training in interdisciplinary research methods.This grant was made as part of the Digging Into Data Challenge, an international competition designed to foster research collaboration across countries and to encourage innovative approaches to analyzing large data sets in the social sciences and humanities. The US based researchers will collaborate with scholars in Canada to achieve the goals of this project.