goru001/nlp-for-marathi

State of the Art Language models and Classifier for Marathi, which is spoken predominantly by Marathi people of Maharashtra, India

NLP for Marathi

This repository contains State of the Art Language models and Classifier for Marathi, which is spoken predominantly by Marathi people of Maharashtra, India.

The models trained here have been used in Natural Language Toolkit for Indic Languages (iNLTK)

Dataset

Created as part of this project

1. Marathi Wikipedia Articles

2. Marathi News Dataset

Open Source Datasets

1. iNLTK Headlines Corpus - Marathi : Uses the Marathi News Dataset prepared above.

Results

Language Model Perplexity (on validation set)

Architecture/Dataset	Marathi Wikipedia Articles
ULMFiT	18
TransformerXL	17.42

Classification Metrics

ULMFiT

Dataset	Accuracy	MCC	Notebook to Reproduce results
iNLTK Headlines Corpus - Marathi	92.40	85.23	Link

Visualizations

Word Embeddings

Architecture	Visualization
ULMFiT	Embeddings projection
TransformerXL	Embeddings projection

Sentence Embeddings

Architecture	Visualization
ULMFiT	Encodings projection

Results of using Transfer Learning + Data Augmentation from iNLTK

On using complete training set (with Transfer learning)

Dataset	Dataset size (train, valid, test)	Accuracy	MCC	Notebook to Reproduce results
iNLTK Headlines Corpus - Marathi	(9672, 1210, 1210)	92.40	85.23	Link

On using 5% of training set (with Transfer learning)

Dataset	Dataset size (train, valid, test)	Accuracy	MCC	Notebook to Reproduce results
iNLTK Headlines Corpus - Marathi	(483, 1210, 1210)	84.13	68.59	Link

On using 5% of training set (with Transfer learning + Data Augmentation)

Dataset	Dataset size (train, valid, test)	Accuracy	MCC	Notebook to Reproduce results
iNLTK Headlines Corpus - Marathi	(483, 1210, 1210)	84.55	69.11	Link

Pretrained Models

Language Models

Download pretrained Language Model from here

Tokenizer

Trained tokenizer using Google's sentencepiece

Download the trained model and vocabulary from here