Common Sense Reasoning as Alternative Scenario Search

常识推理作为替代场景搜索

• 批准号: 519933-2018
• 负责人: Cheung, Jackie
• 金额: $ 0.91万
• 依托单位: McGill University
• 依托单位国家: 加拿大
• 项目类别: Engage Plus Grants Program
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=647493
• 关键词: Common; Sense; Reasoning; Alternative; Scenario

A limitation of current dialogue systems and virtual personal assistant apps is that they can only handle specific**application domains following a more or less standard frame (a.k.a., script or scenario). MSR Maluuba is a**Canadian division of Microsoft that is interested in overcoming this limitation by developing techniques for**performing common sense reasoning. Common sense reasoning involves understanding the typical**characteristics, motivations and intents of people. We have a preliminary system that performs common sense**reasoning, as tested by a task called the Winograd Schema Challenge (WSC). This task involves identifying**entities that are being discussed in certain contexts. In the next phase of the project, we will create an enhanced**dataset of WSC questions, with which we will train a deep learning model for predicting WSC answers. The**outcome of this research will be of use in MSR Maluuba's efforts to develop more flexible and naturalistic**conversational agents.

当前对话系统和虚拟个人助理应用的局限性在于，它们只能遵循或多或少的标准框架（也称为，脚本或场景）。MSR Maluuba是微软在加拿大的一个部门，它有兴趣通过开发执行常识推理的技术来克服这一限制。常识推理涉及了解人们的典型特征、动机和意图。我们有一个执行常识推理的初步系统，通过一个名为Winograd Schema Challenge（WSC）的任务进行了测试。这项任务涉及到识别在特定上下文中正在讨论的 ** 实体。在项目的下一阶段，我们将创建一个增强的WSC问题数据集，我们将用它训练一个深度学习模型来预测WSC答案。这项研究的结果将有助于MSR Maluuba开发更灵活和自然的会话代理。