CASIA at SemEval-2022 Task 11: Chinese Named Entity Recognition for Complex and Ambiguous Entities

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1518 - 1523 July 14-15, 2022 ©2022 Association for Computational Linguistics CASIA at SemEval-2022 Task 11: Chinese Named Entity Recognition for Complex and Ambiguous Entities Jia Fu1,2, Zhen Gan1,3, Zhucong Li1,2, Sirui Li4, Dianbo Sui1,Yubo Chen1,2, Kang Liu1,2, Jun Zhao1,2 1 National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China 2 School of Artificial Intelligence, University of Chinese Academy of Sciences,Beijing, China 3 Beijing University of Chemical Technology, Beijing, China 4 Department of Computer Science, Emory University, Altlanta, GA, USA {zhucong.li, dianbo.sui,yubo.chen, kliu, jzhao}@nlpr.ia.ac.cn, fujia2021@ia.ac.cn,ganzhen@mail.buct.edu.cn, sirui.li@emory.edu Abstract This paper describes our approach to develop a complex named entity recognition system in SemEval 2022 Task 11: MultiCoNER Multilingual Complex Named Entity Recog- nition,Track 9 - Chinese. In this task, we need to identify the entity boundaries and category labels for the six identified categories of CW, LOC, PER, GRP, CORP, and PORD.The task focuses on detecting semantically ambiguous and complex entities in short and low-context settings. We constructed a hybrid system based on Roberta-large model with three training mechanisms and a series of data augmentation. Three training mechanisms include adversarial training, Child-Tuning training, and continued pre-training. The core idea of the hybrid sys- tem is to improve the performance of the model in complex environments by introducing more domain knowledge through data augmentation and continuing pre-training domain adaptation of the model. Our proposed method in this pa- per achieves a macro-F1 of 0.797 on the final test set, ranking second. 1 Introduction SemEval-2022 Task 11: MultiCoNER Multilingual Complex Named Entities Recognition(Malmasi et al., 2022b). This task aims to address the prob- lem of complex and ambiguous named entities in practical and open domain environments . The task has 13 tracks, with track 1 being a multi- lingual track where participants need to train a mul- tilingual model using data from 11 languages. The model should be able to handle monolingual data from any of the languages and code-mixed cases. Tracks 2-12 are monolingual tracks, including En- glish, Spanish, Dutch, Russian, Turkish, Korean, Farsi, German, Chinese, Hindi and Bengali. Par- ticipants are required to train a model for only one language. Track 13 is Code-mixed(Fetahu et al., 2021). This test data contains code-mixed sam- ples. These samples include tokens from any of the 11 mentioned languages in the shared task. We participated in the Chinese monolingual track. Dealing with complex and ambiguous entities in practical and open domain environments is a challenging NLP task(Meng et al., 2021),however, which has not received sufficient attention from the research community. The challenges of this task are mainly in three facts: 1)The complex entities problem, in which complex noun phrases, verbs, in- finitives, or complete sentences, and other entities lack proper nouns, making it difficult to identify them(Ashwini and Choi, 2014).2)The ambiguous entities problem, in which some words are enti- ties in some fields but not in others, especially in search, ASR (Automatic Speech Recognition), and other fields.3)The emerging entities …

Content truncated to 500 words