EAGER: Towards Inter-Sentential Models for Detecting Focus of Negation

理解语言是智能系统的核心，也是许多终端用户应用程序（如机器翻译、问答和文本摘要）所需的组件。大多数从文本中提取意义的计算方法建立了一个语义表示，捕获了人类在阅读文本时直观理解的一些含义。例如，语义角色标记器从纯文本中提取谁对谁做了什么、如何做、何时何地做了什么，语义解析器将文本转换为逻辑形式。这个探索性EAGER项目定义了新的算法，通过检测否定的焦点，从否定语句中提取积极的解释，也就是说，精确定位否定中通常很少的文本块，这些文本块打算被否定。否定是人类语言中普遍存在的一种复杂的语言现象。从理论的角度来看，否定往往具有积极的意义，从隐含到蕴涵不等。尽管观察到这一点，目前理解否定影响的计算方法仅限于范围检测，并且忽略了人类在阅读带有否定元素的文本时立即理解的许多积极解释。该项目建立了句子间模型来检测否定焦点，这是从否定中提取积极解释的关键组成部分。利用否定周围的上下文是改进焦点检测的关键，因为现有的句子内模型未能考虑到从给定否定的许多潜在焦点中发出正确焦点信号的关键相关事件。能够从否定中提取积极的解释将有利于从文本中整体提取意义，因为一旦确定了与否定表达相关的积极解释，它们就可以与对整个文本的解释相结合。

Understanding language is at the core of intelligent systems, and a required component of several end-user applications such as machine translation, question answering and text summarization. Most computational approaches to extract meaning from text build a semantic representation capturing some of the meaning intuitively understood by humans when reading text. For example, semantic role labelers extract who did what to whom, how, when and where from plain text, and semantic parsers transform text into logic forms. This exploratory EAGER project defines new algorithms for extracting positive interpretations from negated statements by detecting the focus of negation, that is, pinpointing the usually few text chunks within a negation that are intended to be negated.Negation is an intricate linguistic phenomenon present in all human languages. From a theoretical point of view, it is well known that negation often carries positive meanings ranging from implicatures to entailments. Despite this observation, current computational approaches to understand the effects of negation are limited to scope detection and disregard the numerous positive interpretations immediately understood by humans when reading text with negated elements. This project builds inter-sentential models to detect the foci of negation, a key component to extract positive interpretations from negations. Exploiting the context surrounding a negation is key to improving foci detection, as existing intra-sentential models fail to take into account key related events that signal the correct foci out of the many potential foci for a given negation. Being able to extract positive interpretations from negation will benefit the overall extraction of meaning from text, since once positive interpretations associated with negative expressions have been identified, they can be integrated with the interpretations obtained for the text as a whole.