CAREER: Understanding Negation in Positive Terms

否定表达是人类语言的一个重要特征。然而，尽管否定无处不在，但当前的智能系统很难理解它的影响，而人们很少这样做。其中最重要的一点是，否定通常传达一些积极的含义。例如，对“约翰从来不会在早上离开家时手里没有咖啡”的自然解释表明，约翰确实在早上离开家，而且他总是手里拿着咖啡。该项目构建了新的资源和算法，使智能系统能够理解隐藏在包含否定的句子中的众多积极解释。这些资源和算法有可能使对话系统能够进行更自然的对话，改进机器翻译，并使几乎任何需要文本理解的最终用户应用程序受益。除了基础研究之外，该项目还将支持博士和本科生的多元化群体，以及北德克萨斯大学的跨学科研讨会，博士生的国际研究停留，研讨会和行业外展活动。该CAREER项目的技术目标分为三个方面。第一个推力创建了一个全面的语料库的否定和他们的积极的解释，包括所有的否定类型在几个领域和体裁。为此，研究团队探索了各种策略，以考虑粒度程度，最大限度地减少注释工作，并对潜在的积极解释进行评分和重写。第二个重点是开发计算算法，从否定中自动生成积极的解释，并学习上下文和话语的表征，然后利用它们来识别否定的焦点，并用积极的对应物重写它们。第三个重点是外部评估，通过将否定纳入到识别蕴涵和推断空间时间线的任务中。该奖项反映了NSF的法定使命，并被认为值得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来支持。

The expression of negation is an important feature of all human languages. However despite the pervasiveness of negation, current intelligent systems have great difficulty comprehending its effects, whereas people rarely do. One of the most important of these is the fact negation generally conveys some positive meaning. For example, a natural interpretation of "John never leaves the house in the morning without a coffee in his hand" suggests that John does leave the house in the morning and that he always does so with a coffee in his hand. This project builds novel resources and algorithms to enable intelligent systems to understand the numerous positive interpretations hidden in sentences containing negation. These resources and algorithms have the potential to empower dialogue systems to carry on more natural conversations, improve machine translation, and benefit virtually any end-user application requiring text understanding. Beyond fundamental research, this project will support a diverse cohort of PhD and undergraduate students as well as an interdisciplinary seminar at University of North Texas, international research stays for PhD students, workshops, and industry outreach activities.The technical aims of this CAREER project are divided into three thrusts. The first thrust creates a comprehensive corpus of negations and their positive interpretations, including all negations types in several domains and genres. To this end, the research team explores strategies to account for degrees of granularity, to minimize the annotation effort and to score and rewrite potential positive interpretations. The second thrust is to develop computational algorithms to automatically generate positive interpretations from negation, and to learn representations of context and discourse, and then leverage them to identify the foci of negation and rewrite them with positive counterparts. The third thrust consists of extrinsic evaluation, by incorporating negation into the tasks of recognizing entailments and inferring spatial timelines.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.