Muril Large Cased

🚀 MuRIL Large

A BERT Large (24L) model pre-trained on 17 Indian languages and their transliterated counterparts, offering multilingual representations for Indian languages.

🚀 Quick Start

This section provides an overview of the MuRIL Large model, including its architecture, training data, and usage scenarios. For more detailed information, please refer to the subsequent sections.

✨ Features

• Multilingual Representation: Pre-trained on 17 Indian languages and their transliterated counterparts, enabling effective processing of multilingual data.
• Modified Training Paradigm: Incorporates translation and transliteration segment pairs in training and uses an exponent value of 0.3 for upsampling to enhance low-resource performance.
• Self-supervised Learning: Trained using a self-supervised masked language modeling task with whole word masking.

📦 Installation

No installation steps were provided in the original README.

💻 Usage Examples

No code examples were provided in the original README.

📚 Documentation

Overview

This model uses a BERT large architecture [1] pretrained from scratch using the Wikipedia [2], Common Crawl [3], PMINDIA [4], and Dakshina [5] corpora for 17 [6] Indian languages.

We use a training paradigm similar to multilingual BERT, with a few modifications as listed:

• We include translation and transliteration segment pairs in training as well.
• We keep an exponent value of 0.3 and not 0.7 for upsampling, shown to enhance low-resource performance. [7]

See the Training section for more details.

Training

The MuRIL model is pre-trained on monolingual segments as well as parallel segments as detailed below:

• Monolingual Data: We make use of publicly available corpora from Wikipedia and Common Crawl for 17 Indian languages.
• Parallel Data: We have two types of parallel data:
  • Translated Data: We obtain translations of the above monolingual corpora using the Google NMT pipeline. We feed translated segment pairs as input. We also make use of the publicly available PMINDIA corpus.
  • Transliterated Data: We obtain transliterations of Wikipedia using the IndicTrans [8] library. We feed transliterated segment pairs as input. We also make use of the publicly available Dakshina dataset.

We keep an exponent value of 0.3 to calculate duplication multiplier values for upsampling of lower resourced languages and set dupe factors accordingly. Note, we limit transliterated pairs to Wikipedia only.

The model was trained using a self-supervised masked language modeling task. We do whole word masking with a maximum of 80 predictions. The model was trained for 1500K steps, with a batch size of 8192, and a max sequence length of 512.

Trainable parameters

All parameters in the module are trainable, and fine-tuning all parameters is the recommended practice.

Uses & Limitations

This model is intended to be used for a variety of downstream NLP tasks for Indian languages. This model is trained on transliterated data as well, a phenomenon commonly observed in the Indian context. This model is not expected to perform well on languages other than the ones used in pre-training, i.e., 17 Indian languages.

Evaluation

We provide the results of fine-tuning this model on a set of downstream tasks. We choose these tasks from the XTREME benchmark, with evaluation done on Indian language test-sets. All results are computed in a zero-shot setting, with English being the high resource training set language. The results for XLM-R (Large) are taken from the XTREME paper [9].

The fine-tuning hyperparameters are as follows:

References

• [1] Jacob Devlin, Ming-Wei Chang, Kenton Lee, Kristina Toutanova. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv preprint arXiv:1810.04805, 2018.
• [2] Wikipedia
• [3] Common Crawl
• [4] PMINDIA
• [5] Dakshina
• [6] Assamese (as), Bengali (bn), English (en), Gujarati (gu), Hindi (hi), Kannada (kn), Kashmiri (ks), Malayalam (ml), Marathi (mr), Nepali (ne), Oriya (or), Punjabi (pa), Sanskrit (sa), Sindhi (sd), Tamil (ta), Telugu (te), and Urdu (ur).
• [7] Conneau, Alexis, et al. Unsupervised cross-lingual representation learning at scale. arXiv preprint arXiv:1911.02116 (2019).
• [8] IndicTrans
• [9] Hu, J., Ruder, S., Siddhant, A., Neubig, G., Firat, O., & Johnson, M. (2020). Xtreme: A massively multilingual multi-task benchmark for evaluating cross-lingual generalization. arXiv preprint arXiv:2003.11080.
• [10] Fang, Y., Wang, S., Gan, Z., Sun, S., & Liu, J. (2020). FILTER: An Enhanced Fusion Method for Cross-lingual Language Understanding. arXiv preprint arXiv:2009.05166.

Citation

If you find MuRIL useful in your applications, please cite the following paper:

@misc{khanuja2021muril,
      title={MuRIL: Multilingual Representations for Indian Languages},
      author={Simran Khanuja and Diksha Bansal and Sarvesh Mehtani and Savya Khosla and Atreyee Dey and Balaji Gopalan and Dilip Kumar Margam and Pooja Aggarwal and Rajiv Teja Nagipogu and Shachi Dave and Shruti Gupta and Subhash Chandra Bose Gali and Vish Subramanian and Partha Talukdar},
      year={2021},
      eprint={2103.10730},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Contact

Please mail your queries/feedback to muril-contact@google.com.