The AI4Bharat-IndicNLP dataset is an ongoing effort to create a collection of large-scale,
general-domain corpora for Indian languages. Currently, it contains 2.7 billion words for 10 Indian languages from two language families.
We share pre-trained word embeddings trained on these corpora.
We create news article category classification datasets for 9 languages to evaluate the embeddings.
We evaluate the IndicNLP embeddings on multiple evaluation tasks.
