AsNER - Annotated Dataset and Baseline for Assamese Named Entity recognition

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6571–6577 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6571 AsNER - Annotated Dataset and Baseline for Assamese Named Entity recognition Dhrubajyoti Pathak, Sukumar Nandi, Priyankoo Sarmah Indian Institute of Technology Guwahati North Guwahati, India {drbj153, sukumar, priyankoo}@iitg.ac.in Abstract We present the AsNER, a named entity annotation dataset for low resource Assamese language with a baseline Assamese NER model. The dataset contains about 99k tokens comprised of text from the speech of the Prime Minister of India and Assamese play. It also contains person names, location names and addresses. The proposed NER dataset is likely to be a significant resource for deep neural based Assamese language processing. We benchmark the dataset by training NER models and evaluating using state-of-the-art architectures for supervised named entity recognition (NER) such as Fasttext, BERT, XLM-R, FLAIR, MuRIL etc. We implement several baseline approaches with state-of-the-art sequence tagging Bi-LSTM-CRF architecture. The highest F1-score among all baselines achieves an accuracy of 80.69% when using MuRIL as a word embedding method. The annotated dataset and the top performing model are made publicly available. Keywords: NER dataset, Language Resources, Assamese NER, Assamese Language, Named Entity Recognition, NER model, AsNER 1. Introduction Named Entity Recognition (NER) aims to clas- sify text in a sentence into predefined classes such as person, location, organization etc. It basically identifies a word or phrase that can be considered as names from a set of documents. NER plays a vital role in preprocessing task of various natu- ral language processing (NLP) applications such as information retrieval (Neudecker, 2016), text un- derstanding (Zhang et al., 2019), automatic text summarization (Larsen, 1999), question answering (Mollá et al., 2006), machine translation (Babych and Hartley, 2003), and knowledge base construc- tion (Etzioni et al., 2005) etc. Recent advances in deep neural network (DL) in NLP exhibit success in various domains. DL-based NER systems with minimal feature engineering have been flourish- ing. Over the past few years, several studies have been reported success in deep learning-based NER model and achieved state-of-the-art performance in resource-rich language (Chiu and Nichols, 2016; Lample et al., 2016; Akbik et al., 2019). The DL- based model requires high quality and large anno- tated datasets for training and evaluation. There- fore, datasets play an essential role in extracting the linguistic features of a language to achieve high performance in downstream tasks. On the other hand, data annotation for a language remains a time consuming and expensive process. It is a major challenge for many resource-poor languages, such as Assamese (Glottocode: assa1263), as it requires language experts to perform annotation tasks on a large amount of data. Suitable anno- tated corpora for Assamese NER, name dictionar- ies, morphological analyzers, dependency parser, POS taggers etc., are not yet publicly available for suitable use in downstream tasks. Although, Assamese has a very old and rich literary history, technology development in NLP is still in a nascent stage. The WikiAnn NER …

Content truncated to 500 words