EAPSI: Learning Semantic Decomposition in Support of Commonsense Reasoning

常识推理是人类使用日常世界事实来支持规划的能力。要想创作一个关于银行抢劫案的故事，人们必须了解银行是什么，为什么想要抢劫它，等等。人工智能(AI)特工可以通过阅读现有的文本来源，如维基百科或叙事，来了解世界很多东西。然而，到目前为止，从这些来源获得的信息往往是显性的，例如，人A出生在地点B。本项目将调查可用于从文本中学习隐含知识的方法，如概念的空间或时间属性。然后，这些知识可以支持人工智能中的许多不同的当前挑战，例如叙事生成。该项目将在丰田工业大学的Makoto Miwa博士的指导下进行。这种协作提供了对独特数据的访问，这些数据将使人们能够对人类如何处理日常概念提供新的见解。目前的信息提取工作倾向于学习明确编码在文本中的命题。隐含信息在赋予认知主体对世界进行推理的能力方面同样重要或更重要。在Miwa博士的指导下，我将把机器学习和类比技术应用到联想概念学习任务中。我将具体研究与低级别质量相关的属性，包括时间性、空间、情感、美学等。我将使用各种语料库来支持这项任务。Miwa博士和他在丰田科技大学的研究小组是信息提取技术的领导者。随后，Miwa博士的洞察力和专业知识将为开发机器学习隐性知识的技术提供无价的帮助。该奖项由东亚和太平洋暑期学院计划设立，支持一名美国研究生的暑期研究，由NSF和日本科学促进会共同资助。

Commonsense reasoning is the ability for humans to use everyday facts about the world to support planning. To create a story about a bank robbery, one must have knowledge of what a bank is, why one would wish to rob it, etc. Artificial intelligence (AI) agents can learn a lot about the world by reading existing text sources, such as Wikipedia or narrative. However, the information learned so far from these sources tends to be explicit, such as, person A was born at location B. This project will investigate methods that can be used to learn implicit knowledge from text such as spatial or temporal properties of concepts. This knowledge can then support many different current challenges in AI, such as narrative generation. This project will be conducted at Toyota Technological University under the mentorship of Dr. Makoto Miwa. The collaboration provides access to unique data that will enable new insights into how humans process everyday concepts.Current work on information extraction tends to learn propositions that are explicitly encoded in text. Implicit information is equally or more important in giving cognitive agents the ability to reason about the world. Under the supervision of Dr. Miwa, I will apply machine learning and analogical techniques to the task of associative concept learning. I will specifically look at low level qualia related attributes including temporality, space, sentiment, aesthetics, etc. I will use various corpora in support of this task. Dr. Miwa, along with his research group at Toyota Technological University, are leaders in information extraction techniques. Subsequently, Dr. Miwa's insight and expertise will be an invaluable aid in developing techniques for machine learning implicit knowledge.This award, under the East Asia and Pacific Summer Institutes program, supports summer research by a U.S. graduate student and is jointly funded by NSF and the Japan Society for the Promotion of Science.