CHS: Small: Automatically Identifying Misunderstandings of Social Intentions in Online Conversations

该项目将开发新的方法来理解社交媒体中的误解，并将设计新的干预措施来防止或减少误解。随着时间的推移，在线对话通常会变得越来越敌意，导致破坏行为，如巨魔、人身攻击、骚扰和欺凌。先前的研究表明，在某些情况下，这些脱轨可以追溯到恶意的个人和社会传染。然而，在其他情况下，干扰可能是简单的误解造成的：读者可能会误解海报的社会意图，得出结论认为信息是友好的、敌对的或咄咄逼人的，而实际上恰恰相反。这个项目将创建一个框架，用于理解社会意图何时以及为什么被误解，计算方法用于自动捕捉这种误解对在线对话的轨迹和结果的影响，以及可能有助于防止误解的工具。这项研究在社会计算和人机交互的背景下，结合了自然语言处理和计算机中介交流的专业知识。与其他研究信息内容如何被误解的工作不同，这项研究的目的是了解识别信息的语用力量的问题，例如，它是指命令或请求、笑话还是敌意言论。这种“语用失误”的根源在于信息的字面意义和意图意义之间的区别，以及不同社会文化背景的读者对两者的不同认知倾向。这种现象在网上尤其有问题，人们依靠线下消除此类话语歧义的许多副语言社交线索被削弱，甚至完全缺失。三组研究活动将：(1)调查信息创建者的语用意图与不同目标受众如何解释它们之间的不匹配；(2)开发技术来自动识别何时发生语用误解并检测其对对话的影响；以及(3)开发接口工具以帮助防止或克服语用失误。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This project will develop new ways of understanding miscommunication in social media and will design novel interventions to prevent or reduce misunderstandings. Online conversations often become increasingly hostile over time, resulting in disruptive behavior such as trolling, personal attacks, harassment, and bullying. Prior work has shown that in some cases these derailments can be traced back to ill-intentioned individuals and to social contagion. In other cases, however, disruptions may result from simple misunderstandings: readers may misconstrue the social intentions of the poster, concluding that a message is meant to be friendly or hostile or aggressive when actually the opposite was intended. This project will create a framework for understanding when and why social intentions are misunderstood, computational methods for automatically capturing the effects of such misperceptions on the trajectory and outcome of online conversations, and tools that might help prevent misunderstandings. The research combines natural language processing and computer-mediated communication expertise, in the context of social computing and human-computer interaction. Unlike other work that examines how the content of messages is misunderstood, this research aims to understand problems identifying the pragmatic force of a message, for example whether it is meant as a command or request, a joke, or a hostile remark. Such "pragmatic failures" are rooted in the distinction between the literal meaning of a message and its intended meaning, as well as in the varying propensity of readers with different sociocultural backgrounds to perceive one or the other. This phenomenon is especially problematic online, where many of the paralinguistic social cues people rely on offline to disambiguate such utterances are attenuated or even completely missing. Three sets of research activities will: (1) investigate mismatches between the message creator's pragmatic intentions and how they are interpreted by different target audiences, (2) develop techniques to automatically identify when pragmatic misunderstandings have occurred and to detect their effects on conversations, and (3) develop interface tools to help prevent or overcome pragmatic failures.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.