dynopii/IndicVarna-100k

IndicVarna for Callchimp.ai (a Dynopii product)

We introduce IndiVarna which was prepared by using Google Translate on the dair-ai/emotion dataset to get the samples there translated to the top 10 most commonly used Indian languages.

This dataset contains 10000 samples of each of the 10 languages supported.

The dataset further translated the labels in the dataset to 3 label sentiments - 0: Negative, 1: Neutral and 2: Positive. Each language has 3334 samples of each category of sentiment.

The dataset has three columns - uuid, text, label. uuid here is the concatenation of uuid with language code separated by -.

This enabled us to start using the dataset out of the box with Hugging Face based text classification pipelines.

The languages supported by the dataset are:

1. English - en
2. Hindi - hi
3. Bengali - bn
4. Gujarati - gu
5. Urdu - ur
6. Kannada - kn
7. Marathi - mr
8. Punjabi - pa
9. Tamil - ta
10. Telugu te

The dataset can be used to perform any of the following downstream tasks -

1. Text classification (sentiment analysis models, similarity models)
2. Text Generation - Fill mask, generation, etc. (samples can be used without the sentiment labels to prepare small corpus for text-generation fine tuning for any of the languages)
3. Translation - every sentence can be matched with corresponding sentences of the same language code and then used to train translations models between the languages.

We look forward to seeing how you use this dataset!

Author

Anubhav Singh (for callchimp.ai)

Contact at: anubhav[at]callchimp.ai for any assistance!