POS-Tagging-for-Hindi

Developing a POS tagger for Hindi Copora

Implementations

• Bi-LSTM Architecture
• Bi-LSTM + CRF Architecture
• Residual LSTM + EMLO (Testing)

TL;DR

1. extract_tags.sh and extract_data are used to extract data from the Hindi Corpora
2. Currently Hindi word embeddings trained on Fasttext are used.
3. train.py files contains the implementation of the above architectures.

Requirements

Requirements:

• Python 3.6
• Keras 2.2.0 - For the creation of BiLSTM-CRF architecture
• Tensorflow 1.8.0 - As backend for Keras (other backends are untested.

Resuts

TEST ACCURACY with Bi-LSTM + CRF : 0.977324