CRI CI-P: Building a Community Resource for Temporal Inference in Chinese

时间推理是识别事件的位置和事件在时间序列中的顺序的任务。为了充分地进行时间推理，必须利用来自各种来源的语言信息。例如，要确定事件的时间位置，需要查看与事件、时态(过去时、现在时)和体(完成时、进行体)标记以及时间状语(`已经)相关的时间表达(`在十点钟)。为了自动进行时间推理，自然语言处理(NLP)领域的主要研究范式是为有监督的机器学习算法提供大量的人类注释数据语料库，这些算法能够结合不同的语言信息源来推断事件的时间位置和顺序。根据规划拨款，私人投资机构获资助与研究界合作，为大型中文语料库制订注解方法，以配合现有的英语资源，支持多语言时态推理的研究。这涉及到从研究界收集以下方面的意见：(1)将现有的英汉时间标注框架扩展到其他亚洲语言；(2)通过现有的语言资源自动引导时间标注的方法。这一领域有影响力的研究人员参加的研讨会是这一规划赠款的核心部分。中文被选为这个项目的主要语言，因为它独特的语言特性，以及缺乏用这种语言进行时间推理所需的语言资源，尽管在这项工作中制定的方法预计很容易推广到其他亚洲语言。时态推理是支持许多自然语言应用的基础技术，包括但不限于信息提取、问答、文本摘要和机器翻译。随着中国崛起为全球强国，先进的时态推理技术将使人们能够以一种具有战略重要性的语言获得更多信息。

Temporal inference is the task of identifying the location of an event and the ordering of events in a temporal sequence. To perform temporal inference adequately, linguistic information from a variety of sources must be exploited. For example, to determine the temporal location of an event, one needs to look at the time expression (``at ten o'clock") associated with the event, tense (past, present) and aspect (perfective, progressive) markers, and temporal adverbials (``already"). To conduct temporal inference automatically, the dominant research paradigm in the Natural Language Processing (NLP) community is to provide a large corpus of human annotated data for supervised machine learning algorithms, which are capable of combining the divergent sources of linguistic information to infer the temporal location and order of events. Under the planning grant, the PIs are funded to work with the research community to define an annotation methodology for a large-scale Chinese corpus that can support research in multilingual temporal inference in conjunction with existing resources for English. This involves gathering input from the research community on (1) the extension of the existing temporal annotation framework for English to Chinese and possibly to other Asian languages, (2) the methods to bootstrap the temporal annotation automatically through existing linguistic resources. A workshop attended by influential researchers in this area is a central part of this planning grant. Chinese is chosen as the primary language for this project because of its distinctive linguistic properties and the lack of necessary linguistic resources to perform temporal inference in this language, although the methodologies developed in this effort are expected to be readily generalizable to other Asian languages. Temporal inference is a fundamental technology that supports many natural language applications, including but not limited to Information Extraction, Question Answering, Text Summarization and Machine Translation. With the rise of China as a global power, advancing temporal inference technology will enable greater information access in a language that has strategic importance.