Awesome

• Practical Comparable Data Collection for Low-Resource Languages via Images
• Source
• Data
• Task Instructions
• Citation

Practical Comparable Data Collection for Low-Resource Languages via Images

• Paper: https://arxiv.org/abs/2004.11954

• Slides: https://madaan.github.io/res/artifacts/pml4dc-practical-data-collection.pdf

• We propose a method of curating high-quality comparable training data for low-resource languages without requiring that the annotators are bilingual. Our method involves using a carefully selected set of images as a pivot between the source and target languages by getting captions for such images in both languages independently. Human evaluations on the English-Hindi comparable corpora created with our method show that 81.1% of the pairs are acceptable translations, and only 2.47% of the pairs are not a translation at all. We further establish the potential of dataset collected through our approach by experimenting on two downstream tasks -- machine translation and dictionary extraction.

• Joint work with Shruti Rijhwani, Antonios Anastasopoulos, Yiming Yang, and Graham Neubig.

• Work to be presented as a poster at Practical ML for Developing Countries Workshop at ICLR 2020.

Source

• Translation Downstream Task - (Paul Michel's Baseline NMT Implementation)[https://github.com/pmichel31415/11731-assignment-2-baseline]

• Unsupervised Dictionary Extraction

python src/make_dict.py -i data/alignments/wys.fastalign.input -a data/alignments/symmetric.align -l1 hin -l2 eng 

• Extracting Counts

python src/find_token_alignments.py data/alignments/wys.fastalign.input data/alignments/symmetric.align output_path

Data

data/
├── alignments
│   ├── forward.align
│   ├── reverse.align
│   ├── symmetric.align
│   └── wys.fastalign.input
├── captions.tsv
└── dict.hin-eng.txt

Task Instructions

• आपको एक वाक्य के साथ हर छवि का वर्णन करना होगा।
• छवि के वर्णन में शुद्ध हिंदी का उपयोग करना ज़रूरी नहीं हे | जैसा ठीक लगे लिखें |
• कृपया उन गतिविधियों, लोगों, जानवरों और वस्तुओं का सटीक विवरण प्रदान करें जिन्हें आप चित्र में देख रहे हें |
• प्रत्येक विवरण एक ही वाक्य का होना चाहिए।
• विवरण हिंदी में लिखा जाना चाहिए।
• संक्षिप्त होने का प्रयास करें।
• अगर किसी शब्द का हिंदी में मतलब ना पता हो तो उसे इंग्लिश में ही लिख दें |
• व्याकरण और वर्तनी पर ध्यान दें।

Citation

If you use our work, please cite:

@inproceedings{madaan2020practical,
  title={Practical Comparable Data Collection for Low-Resource Languages via Images},
  author={Madaan, Aman and Rijhwani, Shruti and Anastasopoulos, Antonios and Yang, Yiming and Neubig, Graham},
  booktitle={Proceedings of the Practical ML for Developing Countries Workshop, ICLR 2020},
  year={2020}
}