SLPG's picture
Update README.md
b5b4b19 verified
metadata
datasets:
  - SLPG/Punjabi_Transliteration_Corpus
language:
  - pa
metrics:
  - bleu
  - cer
library_name: fairseq
pipeline_tag: translation
tags:
  - punjabi shahmukhi
  - punjabi gurmukhi
  - transliteration
  - punjabi transliteration
  - punjabi gur to shahmukhi
  - transliteration system
  - punjabi transliteration system

Punjabi Gurmukhi to Shahmukhi Transliteration System

Our supervised Punjabi transliteration systems built using unsupervised corpus are bidirectional NMT systems which effectively convert text between Gurmukhi and Shahmukhi scripts. The Gurmukhi-to-Shahmukhi model achieves a 98.1 BLEU score and 99.5% word-level accuracy, while the Shahmukhi-to-Gurmukhi model scores 87.7 BLEU.

Corpus Details

  • Total Sentences: 6.3 million
  • Domains Covered: Various domains including CCaligned, ccmatrix, TED, QED, OPUS, TIco, Wikimedia, Multicclaigned, Emille, IJCNLP, xlent, and paracrawl.
  • Test Corpus: FLORES-101

Model Details

- **BLEU Score:** 98.1
- **Word-level Accuracy:** 99.5%
- **Character Error Rate (CER):** 99.1%

You may also explore our Shahmukhi-to-Gurmukhi Model with BLEU Score of 87.7 here.

Usage

These resources are intended to facilitate research and development in the field of Punjabi transliteration. They can be used to train new models or improve existing ones, enabling high-quality transliteration between Gurmukhi and Shahmukhi scripts.

Citation

If you use our model, kindly cite our paper:

@article{Shehzadi2024,
  title={Unsupervised Punjabi Corpus and Neural Machine Transliteration
 System},
  author={Shehzadi Ambreen, Sadaf Abdul Rauf, MG Abbas Malik and Muhammad Imran },      journal={Heliyon},
  year={2024},
  note={Under review}
 }