Naamapadam: A Large-Scale Named Entity Annotated Data for Indic Languages

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 10441–10456 July 9-14, 2023 ©2023 Association for Computational Linguistics Naamapadam: A Large-Scale Named Entity Annotated Data for Indic Languages Arnav Mhaske1,2∗ Harshit Kedia1,2∗ Sumanth Doddapaneni1,2 Mitesh M. Khapra1,2 Pratyush Kumar1,2,3 Rudra Murthy V2,4† Anoop Kunchukuttan1,2,3† 1Indian Institute of Technology Madras 2AI4Bharat 3Microsoft India 4IBM Research India Abstract We present, Naamapadam, the largest publicly available Named Entity Recognition (NER) dataset for the 11 major Indian languages from two language families. The dataset contains more than 400k sentences annotated with a to- tal of at least 100k entities from three standard entity categories (Person, Location, and, Orga- nization) for 9 out of the 11 languages. The training dataset has been automatically created from the Samanantar parallel corpus by pro- jecting automatically tagged entities from an English sentence to the corresponding Indian language translation. We also create manually annotated testsets for 9 languages. We demon- strate the utility of the obtained dataset on the Naamapadam-test dataset. We also release In- dicNER, a multilingual IndicBERT model fine- tuned on Naamapadam training set. IndicNER achieves an F1 score of more than 80 for 7 out of 9 test languages. The dataset and models are available under open-source licences at https: //ai4bharat.iitm.ac.in/naamapadam. 1 Introduction Named Entity Recognition (NER) is a fundamen- tal task in natural language processing (NLP) and is an important component for many downstream tasks like information extraction, machine trans- lation, entity linking, co-reference resolution, etc. The most common entities of interest are person, location, and, organization names, which are the focus of this work and most work in NLP. Given high-quality NER data, it is possible to train good- quality NER systems with existing technologies (Devlin et al., 2019). For many high-resource lan- guages, publicly available annotated NER datasets (Tjong Kim Sang, 2002; Tjong Kim Sang and De Meulder, 2003a; Pradhan et al., 2013; Benikova et al., 2014) as well as high-quality taggers (Wang et al., 2021; Li et al., 2020) are available. ∗Equal contribution †Project leads; correspondence to rmurthyv@in.ibm.com, Figure 1: Illustration of Named Entity projection. We perform (i) NER with fine-tuned English BERT model, followed by (ii) word alignment between parallel sen- tence pair and (iii) projection of the English entities onto Indic sentence. However, most Indic languages do not have suffi- cient labeled NER data to build good-quality NER models. All existing NER corpora for Indic lan- guages have been manually curated (Lalitha Devi et al., 2014; Murthy et al., 2018; Pathak et al., 2022; Murthy et al., 2022; Malmasi et al., 2022; Litake et al., 2022). Given the number of languages, the expenses and the logistical challenges, these datasets are limited on various dimensions viz. cor- pus size, language coverage, and broad domain representation. In recent years, zero-shot cross- lingual transfer from pre-trained models, fine-tuned on task-specific training data in English has been proposed as a way to support various language un- derstanding tasks for low-resource languages (Hu et al., 2020). However, …

Content truncated to 500 words