Kannada Named Entity Recognition and Classification using Bidirectional Long Short-Term Memory Networks

This paper focuses on carrying out the Named Entity Recognition and Classification (NERC) task on Kannada, a major Dravidian language spoken in India. Low resource conditions such as absence of external linguistic resources and gazetteers in Kannada and other Dravidian languages pose obstacles to the NERC task. LSTM networks, with their capability of learning long-term dependencies, present an effective solution to this task without the need of a deeper understanding of the semantics of the language. This paper describes a novel supervised machine learning model for Kannada NERC using Bidirectional LSTM networks. The network model is trained and validated on a manually annotated corpus, and gives encouraging results in terms of various evaluation metrics.