Qtrade AI at SemEval-2022 Task 11: An Unified Framework for Multilingual NER Task

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1654 - 1664 July 14-15, 2022 ©2022 Association for Computational Linguistics Qtrade AI at SemEval-2022 Task 11: An Unified Framework for Multilingual NER Task Weichao Gan∗†, Yuanping Lin∗†, Guangbo Yu∗†, Guimin Chen † and Qian Ye § †Qtrade AI Lab, Shenzhen China §Guangdong Power Grid Corporation, Guangdong China {waiciugam, yuenpinglynn, ygbusc}@gmail.com chenguimin@foxmail.com, eason_sysuye@163.com Abstract This paper describes our system, which placed third in the Multilingual Track (subtask 11), fourth in the Code-Mixed Track (subtask 12), and seventh in the Chinese Track (subtask 9) in the SemEval 2022 Task 11: MultiCoNER Multilingual Complex Named Entity Recog- nition. Our system’s key contributions are as follows: 1) For multilingual NER tasks, we of- fer an unified framework with which one can easily execute single-language or multilingual NER tasks, 2) for low-resource code-mixed NER task, one can easily enhance his or her dataset through implementing several simple data augmentation methods and 3) for Chinese tasks, we propose a model that can capture Chinese lexical semantic, lexical border, and lexical graph structural information. Finally, our system achieves macro-f1 scores of 77.66, 84.35, and 74.00 on subtasks 11, 12, and 9, respectively, during the testing phase. 1 Introduction SemEval 2022 Task 11: MultiCoNER Multilin- gual Complex Named Entity Recognition(Malmasi et al., 2022b) focuses on extracting semantically ambiguous complex named entities(Meng et al., 2021) in short, low-context and code-mixed(Fetahu et al., 2021) scenarios. The domain adaptabil- ity capacity of the system is in high demand for this shared task, which contains 13 tracks in En- glish, Spanish, Dutch, Russian, Turkish, Korean, Farsi, German, Chinese, Hindi, Bangla, multi- language, and code-mixed. Among them, multi- language, and code-mixed tracks are in the mixed language, while the other tracks are in the single language. The difference between multi-language track and code-mixed track is that the corpus of multi-language track is multilingual, but the words in each sentence are in a single language, while the corpus of code-mixed track is a corpus in which the *These authors contributed equally to this work. words or phrases in each sentence may be from dif- ferent languages, for example the sentence I Liebe Sie (I love you) consists of English, French and Ger- man. Furthermore, the datasets offered by the or- ganizers , which may mainly from Wikipedia, web questions and user queries, comprise data from 11 languages, with each language containing around 15,000 training samples and 800 development (dev) samples(Malmasi et al., 2022a). The corpus con- tains a total of six entity categories, namely loca- tion (LOC), person (PER), product (PROD), group (GRP), corporation (CORP) and creative works (CW). The main contributions of our system are as follows: 1) we propose a unified framework for multilingual NER tasks, using which one can easily perform monolingual or multilingual NER tasks; 2) for low-resource code-mixed NER tasks, we pro- vide several simple and effective data augmentation methods to easily increase the amount of data; 3) for Chinese tasks, we propose a model that captures …

Content truncated to 500 words