针对现有流水线式事件抽取方法依靠大量训练数据、在低资源情况下难以快速迁移运用等问题,利用提示调优技术,提出适用于低资源场景下的流水线式事件抽取方法(low-resource event extraction method using the multi-information fusion ...针对现有流水线式事件抽取方法依靠大量训练数据、在低资源情况下难以快速迁移运用等问题,利用提示调优技术,提出适用于低资源场景下的流水线式事件抽取方法(low-resource event extraction method using the multi-information fusion with prompt tuning,IFPT)。该方法通过构造语义映射和提示模板充分利用事件类型描述、实体类型等多种信息,能够高效使用有限训练数据,流水线式地完成事件检测和论元抽取。实验结果表明,在低资源情况下,IFPT方法论元抽取性能超过了所有基准模型,采取流水线方式能够达到与SOTA模型相媲美的性能。展开更多
In this paper,the static output feedback stabilization for large-scale unstable second-order singular systems is investigated.First,the upper bound of all unstable eigenvalues of second-order singular systems is deriv...In this paper,the static output feedback stabilization for large-scale unstable second-order singular systems is investigated.First,the upper bound of all unstable eigenvalues of second-order singular systems is derived.Then,by using the argument principle,a computable stability criterion is proposed to check the stability of secondorder singular systems.Furthermore,by applying model reduction methods to original systems,a static output feedback design algorithm for stabilizing second-order singular systems is presented.A simulation example is provided to illustrate the effectiveness of the design algorithm.展开更多
As a subtask of information extraction (IE), which aims to extract structured information from texts, event extraction is to recognize event trigger mentions of a predefined event type and their arguments. In general,...As a subtask of information extraction (IE), which aims to extract structured information from texts, event extraction is to recognize event trigger mentions of a predefined event type and their arguments. In general, event extraction can be divided into two subtasks: trigger extraction and argument extraction. Currently, the frequent existences of unannotated trigger mentions and poor-context trigger mentions impose critical challenges in Chinese trigger extraction. This paper proposes a novel three-layer joint model to integrate three components in trigger extraction, i.e., trigger identification, event type determination, and event subtype determination. In this way, different kinds of evidence on distinct pseudo samples can be well captured to eliminate the harmful effects of those un-annotated trigger mentions. In addition, this paper introduces various types of linguistically driven constraints on the trigger and argument semantics into the joint model to recover those poor-context trigger mentions. The experimental results show that our joint model significantly outperforms the state-of-the-art Chinese trigger extraction and Chinese event extraction as a whole.展开更多
文摘针对现有流水线式事件抽取方法依靠大量训练数据、在低资源情况下难以快速迁移运用等问题,利用提示调优技术,提出适用于低资源场景下的流水线式事件抽取方法(low-resource event extraction method using the multi-information fusion with prompt tuning,IFPT)。该方法通过构造语义映射和提示模板充分利用事件类型描述、实体类型等多种信息,能够高效使用有限训练数据,流水线式地完成事件检测和论元抽取。实验结果表明,在低资源情况下,IFPT方法论元抽取性能超过了所有基准模型,采取流水线方式能够达到与SOTA模型相媲美的性能。
基金Project supported by the National Natural Science Foundation of China(Nos.11971303 and 11871330)。
文摘In this paper,the static output feedback stabilization for large-scale unstable second-order singular systems is investigated.First,the upper bound of all unstable eigenvalues of second-order singular systems is derived.Then,by using the argument principle,a computable stability criterion is proposed to check the stability of secondorder singular systems.Furthermore,by applying model reduction methods to original systems,a static output feedback design algorithm for stabilizing second-order singular systems is presented.A simulation example is provided to illustrate the effectiveness of the design algorithm.
文摘As a subtask of information extraction (IE), which aims to extract structured information from texts, event extraction is to recognize event trigger mentions of a predefined event type and their arguments. In general, event extraction can be divided into two subtasks: trigger extraction and argument extraction. Currently, the frequent existences of unannotated trigger mentions and poor-context trigger mentions impose critical challenges in Chinese trigger extraction. This paper proposes a novel three-layer joint model to integrate three components in trigger extraction, i.e., trigger identification, event type determination, and event subtype determination. In this way, different kinds of evidence on distinct pseudo samples can be well captured to eliminate the harmful effects of those un-annotated trigger mentions. In addition, this paper introduces various types of linguistically driven constraints on the trigger and argument semantics into the joint model to recover those poor-context trigger mentions. The experimental results show that our joint model significantly outperforms the state-of-the-art Chinese trigger extraction and Chinese event extraction as a whole.
