An HMM Based Named Entity Recognition System for Indian Languages: The JU System at ICON 2013

TL;DR

This paper presents an HMM-based named entity recognition system for multiple Indian languages, evaluated on ICON 2013 datasets, achieving varying levels of accuracy across languages.

Contribution

Introduces a statistical HMM approach for NER in Indian languages and reports its performance on standard datasets.

Findings

Achieved high F-measure for Bengali and English.

Lower performance on Tamil and Telugu.

Demonstrates effectiveness of HMM for multilingual NER.

Abstract

This paper reports about our work in the ICON 2013 NLP TOOLS CONTEST on Named Entity Recognition. We submitted runs for Bengali, English, Hindi, Marathi, Punjabi, Tamil and Telugu. A statistical HMM (Hidden Markov Models) based model has been used to implement our system. The system has been trained and tested on the NLP TOOLS CONTEST: ICON 2013 datasets. Our system obtains F-measures of 0.8599, 0.7704, 0.7520, 0.4289, 0.5455, 0.4466, and 0.4003 for Bengali, English, Hindi, Marathi, Punjabi, Tamil and Telugu respectively.