USTC-NELSLIP at SemEval-2022 Task 11: Gazetteer-Adapted Integration Network for Multilingual Complex Named Entity Recognition

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1613 - 1622 July 14-15, 2022 ©2022 Association for Computational Linguistics USTC-NELSLIP at SemEval-2022 Task 11: Gazetteer-Adapted Integration Network for Multilingual Complex Named Entity Recognition Beiduo Chen1, Jun-Yu Ma1, Jiajun Qi1, Wu Guo1, Zhen-Hua Ling1 and Quan Liu2 1National Engineering Laboratory for Speech and Language Information Processing, University of Science and Technology of China 2State Key Laboratory of Cognitive Intelligence, iFLYTEK Research {beiduo,mjy1999,jiajun97}@mail.ustc.edu.cn {guowu,zhling}@ustc.edu.cn, quanliu@iflytek.com Abstract This paper describes the system developed by the USTC-NELSLIP team for SemEval-2022 Task 11 Multilingual Complex Named Entity Recognition (MultiCoNER). We propose a gazetteer-adapted integration network (GAIN) to improve the performance of language mod- els for recognizing complex named entities. The method first adapts the representations of gazetteer networks to those of language models by minimizing the KL divergence between them. After adaptation, these two networks are then integrated for backend supervised named entity recognition (NER) training. The proposed method is applied to several state-of- the-art Transformer-based NER models with a gazetteer built from Wikidata, and shows great generalization ability across them. The final predictions are derived from an ensemble of these trained models. Experimental results and detailed analysis verify the effectiveness of the proposed method. The official results show that our system ranked 1st on three tracks (Chinese, Code-mixed and Bangla) and 2nd on the other ten tracks in this task. 1 Introduction Named Entity Recognition (NER) is a core natural language processing (NLP) task, which aims at finding entities and recognizing their type in a text sequence. In practical and open-domain settings, it is difficult for machines to process complex and ambiguous named entities (Ashwini and Choi, 2014). For example, “On the Beach” is the title of a movie but cannot be recognized easily by present NER systems. This issue may become even more serious in multilingual or code-mixed settings (Fetahu et al., 2021). However, it has not received sufficient attention from the research community. To alleviate the issue, SemEval-2022 Task 11 (Malmasi et al., 2022b) formulates this task which focuses on detecting semantically ambiguous and complex entities in short and low-context settings for 11 languages. One of the classic approaches to solving this problem is to integrate external entity knowledge or gazetteers into neural architectures (Liu et al., 2019; Rijhwani et al., 2020; Meng et al., 2021). Typically, the two representations respectively from a language model like BERT (Devlin et al., 2019) and a gazetteer network like BiLSTM (Hochreiter and Schmidhuber, 1997) are combined as one merged embedding, which is further fed into a NER classifier such as a conditional random field (CRF) (Lafferty et al., 2001). However, there is a sense of “gap” between the two networks. The gazetteer network has no explicit semantic learning goal itself, which means it is just a more complex but almost isolated embedding layer for gazetteer information and cannot obtain the true meaning of NER tags actively. Almost no semantic information can be gained by the classic gazetteer network. To effectively connect the …

Content truncated to 500 words