V\=arta: A Large-Scale Headline-Generation Dataset for Indic Languages

TL;DR

Varta is a comprehensive large-scale dataset of 41.8 million news articles across 14 Indic languages, enabling advancements in multilingual NLP and headline generation, and serving as a resource for pretraining language models.

Contribution

The paper introduces Varta, the largest curated Indic language news dataset, and demonstrates its utility for training and evaluating multilingual NLP models.

Findings

State-of-the-art models struggle with the dataset's complexity.

Models perform only slightly better than extractive baselines.

Pretraining on Varta improves performance on NLU and NLG benchmarks.

Abstract

We present V\=arta, a large-scale multilingual dataset for headline generation in Indic languages. This dataset includes 41.8 million news articles in 14 different Indic languages (and English), which come from a variety of high-quality sources. To the best of our knowledge, this is the largest collection of curated articles for Indic languages currently available. We use the data collected in a series of experiments to answer important questions related to Indic NLP and multilinguality research in general. We show that the dataset is challenging even for state-of-the-art abstractive models and that they perform only slightly better than extractive baselines. Owing to its size, we also show that the dataset can be used to pretrain strong language models that outperform competitive baselines in both NLU and NLG benchmarks.