Abstract
We present iNLTK, an open-source NLP library consisting of pre-trained language models and out-of-the-box support for Paraphrase Generation, Textual Similarity, Sentence Embeddings, Word Embeddings, Tokenization and Text Generation in 13 Indic Languages. By using pre-trained models from iNLTK for text classification on publicly available datasets, we significantly outperform previously reported results. On these datasets, we also show that by using pre-trained models and paraphrases from iNLTK, we can achieve more than 95% of the previous best performance by using less than 10% of the training data. iNLTK is already being widely used by the community and has 40,000+ downloads, 600+ stars and 100+ forks on GitHub. The library is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2009.12534
