CSECU-DSG at SemEval-2022 Task 11: Identifying the Multilingual Complex Named Entity in Text Using Stacked Embeddings and Transformer based Approach

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1549 - 1555 July 14-15, 2022 ©2022 Association for Computational Linguistics CSECU-DSG at SemEval-2022 Task 11: Identifying the Multilingual Complex Named Entity in Text Using Stacked Embeddings and Transformer based Approach Abdul Aziz, Md. Akram Hossain, and Abu Nowshed Chy Department of Computer Science and Engineering University of Chittagong, Chattogram-4331, Bangladesh {aziz.abdul.cu, akram.hossain.cse.cu}@gmail.com, and nowshed@cu.ac.bd Abstract Recognizing complex and ambiguous named entities (NEs) is one of the formidable tasks in the NLP domain. However, the diversity of linguistic constituents, syntactic structure, se- mantic ambiguity as well as differences from traditional NEs make it challenging to identify the complex NEs. To address these challenges, SemEval-2022 Task 11 introduced a shared task MultiCoNER focusing on complex named en- tity recognition in multilingual settings. This paper presents our participation in this task where we propose two different approaches in- cluding a BiLSTM-CRF model with stacked- embedding strategy and a transformer-based approach. Our proposed method achieved com- petitive performance among the participants’ methods in a few languages. 1 Introduction Named entity recognition (NER) is a popular se- quence labeling task in the natural language pro- cessing (NLP) arena. It has numerous applications in several computational linguistic tasks including designing efficient search systems, data mining, and document indexing. However, prior studies mostly focused on identifying traditional named entities (NEs) recognition e.g. person names, lo- cations, and organizations names (Murthy et al., 2018). The ever-growing generation of unstructured so- cial media data contains a huge amount of com- plex (Meng et al., 2021) and ambiguous(Fetahu et al., 2021) named entities. This is because so- cial media data is severely induced by noise as well as the linguistic constituent, syntactic and se- mantic ambiguity exists in this data source. Be- sides, the social media data mostly have multilin- gual data. Therefore it poses new challenges to the traditional named entity recognizer system (Aguilar **The first two authors have equal contributions. et al., 2019; Ashwini and Choi, 2014). To ad- dress the challenges of recognizing such complex and ambiguous named entities in multilingual set- tings, (Malmasi et al., 2022b) introduced a shared task at SemEval-2022 named as MultiCoNER. The task is composed of three category of tracks includ- ing multi-lingual, mono-lingual, and code-mixed tracks. A multilingual dataset (Malmasi et al., 2022a) containing data from 11 languages is used to assess the participants’ system. To illustrate a clear view of the task definition, we articulate two examples from English languages and correspond- ing labels in Table 1. English Text#1: Adaptation of seinen series by kenichi sonoda. Tag: [O, O, B-CW, O, O, B-PER, I-PER, O] Text#2: It was designed by kohn pedersen fox. Tag: [O, O, O, O, B-CORP, I-CORP, I-CORP, O ] Table 1: Data sample. We articulate the rest of the contents as fol- lows: Section 2 describes our proposed approach whereas, in Section 3, we present our experimental setup and conduct performance analysis against the various settings and participants’ methods. Finally, we conclude our work …

Content truncated to 500 words