CHED: A Cross-Historical Dataset with a Logical Event Schema for Classical Chinese Event Detection

CHED: A Cross-Historical Dataset with a Logical Event Schema for Classical Chinese Event Detection Congcong Wei ∗, Zhenbing Feng ∗, Shutan Huang, Wei Li, Yanqiu Shao† Information Science School, Beijing Language and Culture University Language Resources Monitoring and Research Center 15 Xueyuan Road, HaiDian District, Beijing, 100083 weicongcong0214@163.com, zbfengblcu@163.com liweitj47@blcu.edu.cn, yqshao163@163.com Abstract Event detection (ED) is a crucial area of natural language processing that automates the extrac- tion of specific event types from large-scale text, and studying historical ED in classical Chinese texts helps preserve and inherit historical and cultural heritage by extracting valuable informa- tion. However, classical Chinese language characteristics, such as ambiguous word classes and complex semantics, have posed challenges and led to a lack of datasets and limited research on event schema construction. In addition, large-scale datasets in English and modern Chinese are not directly applicable to historical ED in classical Chinese. To address these issues, we con- structed a logical event schema for classical Chinese historical texts and annotated the resulting dataset, which is called classical Chinese Historical Event Dataset (CHED). The main challenges in our work on classical Chinese historical ED are accurately identifying and classifying events within cultural and linguistic contexts and addressing ambiguity resulting from multiple mean- ings of words in historical texts. Therefore, we have developed a set of annotation guidelines and provided annotators with an objective reference translation. The average Kappa coefficient after multiple cross-validation is 68.49%, indicating high quality and consistency. We conducted various tasks and comparative experiments on established baseline models for historical ED in classical Chinese. The results showed that BERT+CRF had the best performance on sequence labeling task, with an f1-score of 76.10%, indicating potential for further improvement. 1 1 Introduction Event detection (ED) is a significant research area in natural language processing (NLP). The ED task mainly includes two steps. Firstly, recognizing and labeling triggers (words that best represent the oc- currence of events) in the text, and secondly, determining the event types to which triggers belongs. For example, in the sentence “九月乙丑，太尉李修罢。” (In September of Yi Chou, General Li Xiu was dismissed.), the word “罢” (ba) means “dismiss”. Therefore, the trigger in this sentence is “罢” (ba), and we label this sentence as a “职位-官位-免职” (Position-Official position-Dismiss from a position) event triggered by the word “罢” (ba). Constructing high-quality datasets for specific domains is critical for ED tasks. Several high-quality ED datasets exist for English and Chinese, such as ACE 2005 (Walker et al., 2006), LEVEN (Yao et al., 2022), MAVEN (Wang et al., 2020), PoE (Li et al., 2022) and DuEE (Li et al., 2020). However, classical Chinese lacks such datasets due to complex semantics and special era. Large-scale datasets in English and modern Chinese are not directly applicable to classical Chinese ED. The current research on ED in classical Chinese is limited by the lack of high-quality datasets that are specific, systematic, and scalable. To address these crucial issues and enhance the accuracy and efficiency of classical Chinese ED, we have constructed the …

Content truncated to 500 words