Raccoons at SemEval-2022 Task 11: Leveraging Concatenated Word Embeddings for Named Entity Recognition

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1576 - 1582 July 14-15, 2022 ©2022 Association for Computational Linguistics Raccoons at SemEval-2022 Task 11: Leveraging Concatenated Word embeddings for Named Entity Recognition Atharvan Dogra Thapar University, Patiala, India atharvandogra007@gmail.com Prabsimran Kaur Thapar University, Patiala, India pkaur_be18@thapar.edu Guneet Singh Kohli Thapar University, Patiala, India guneetsk99@gmail.com Jatin Bedi Thapar University, Patiala, India jatin.bedi@thapar.edu Abstract Named Entity Recognition (NER), an essential subtask in NLP that identifies text belonging to predefined semantics such as a person, lo- cation, organization, drug, time, clinical pro- cedure, biological protein, etc. NER plays a vital role in various fields such as information extraction, question answering, and machine translation. This paper describes our participat- ing system run to the Named entity recognition and classification shared task SemEval-2022. The task is motivated towards detecting seman- tically ambiguous and complex entities in short and low-context settings. Our team focused on improving entity recognition by improving the word embeddings. We concatenated the word representations from State-of-the-art language models and passed them to find the best repre- sentation through a reinforcement trainer. Our results highlight the improvements achieved by various embedding concatenations. 1 Introduction Named entity recognition and classification is an essential subtask in natural language processing (NLP). The task of identifying named entities like a person, location, organization, drug, time, clin- ical procedure, biological protein, etc. in a text document is key to many applications, including information extraction and retrieval, machine trans- lation(Al-Onaizan and Knight, 2002; Steinberger et al., 2013), topic modeling(Newman et al., 2006), text summarization(Schiffman et al., 2002), and question answering. It can also be used in domain- specific entity types such as disease, symptoms, and treatments. Recently many researchers have identified the name identification issue in a variety of languages and have made efforts to apply named entity recognition. Complex and ambiguous Named entities, like the title of creative works such as books, songs, and movies, are not simple nouns and are thus harder to recognize(Ashwini and Choi, 2014). Unlike the traditional NEs these works take the form of lin- guistic constituents such as "Remember me when we are parted," which is an imperative clause and does not look like the traditional NEs (Locations, Person name, organizations). This ambiguity in syntax makes it difficult to identify them based on their context. There can also occur instances where these titles are semantically ambiguous; for exam- ple, "the girl on the train" can be a preposition or the name of a book. Such entities are growing at a faster rate as compared to the traditional categories. Therefore, processing these NEs has always been a challenging task in NLP in practical and open- domain settings. However, it has received enough engagement from the research community. ralDespite the high score produced by Neu- models on benchmark datasets like CoNLL03/OntoNotes, it has been noticed (Augenstein et al., 2017) that these models perform significantly low on complex or unseen data. This happens because the scores were driven by the presence of easy …

Content truncated to 500 words