SMUQamar/Kashmiri-English-Parallel-Corpus

Kashmiri-English Parallel Corpus

This dataset is a parallel corpus of Kashmiri and English, consisting of 30,000 sentence pairs(also contains filtered already available corpus). The original dataset contains 270,000 sentence pairs, which are available upon request. This dataset can be utilized for various NLP tasks, including machine translation, alignment studies, and linguistic research.

Corpus Structure

The dataset is organized into several directories to differentiate between raw, cleaned, and processed data:

Raw Data: This contains the unprocessed, original text files in Kashmiri and English. The raw data includes all original content, including extra spaces, punctuation, or any other noise that has not been filtered out.

Cleaned Data: These files have been preprocessed to remove extra spaces, punctuation, or any kind of noise. This version is more suitable for immediate use in NLP tasks where clean data is required.

Processed Data: JSON files containing the aligned Kashmiri-English sentence pairs, formatted for use in machine translation and other NLP applications.

Corpus Details

Size: 30,000 sentence pairs (270,000 pairs available upon request) Sources: The data is sourced from three different origins: Filtered from existing data Manually created OCR scanned

Dataset Structure

The dataset is organized into the following structure: HuggingFace 30K/ ├── Raw Data/ │ ├── Filtered(R)/ │ │ ├── English.txt │ │ └── Kashmiri.txt │ ├── Manual(R)/ │ │ ├── English.txt │ │ └── Kashmiri.txt │ └── OCR Scanned(R)/ │ ├── English.txt │ └── Kashmiri.txt ├── Cleaned Data/ │ ├── Combined Cleaned/ │ │ ├── English.txt │ │ └── Kashmiri.txt │ ├── Filtered(C)/ │ │ ├── English_cleaned.txt │ │ └── Kashmiri_cleaned.txt │ ├── Manual(C)/ │ │ ├── English_cleaned.txt │ │ └── Kashmiri_cleaned.txt │ └── OCR Scanned(C)/ │ ├── English_cleaned.txt │ └── Kashmiri_cleaned.txt ├── Processed Corpus/ │ ├── en_ks_full.jsonl │ └── ks_en_full.jsonl

Citation

If you use this dataset in your research, please cite it as follows: Qumar, S.M.U., Azim, M. & Quadri, S.M.K. Addressing the data gap: building a parallel corpus for Kashmiri language. Int. j. inf. tecnol. (2024). https://doi.org/10.1007/s41870-024-01979-8

Data Dictionary

The data_dictionary.json file included in this repository provides a detailed description of the structure and meaning of each field within the processed JSON files. This is essential for understanding how to use the dataset effectively in NLP tasks.

Example Usage

An example script (example_usage.py) are provided in this repository to demonstrate how to load, preprocess, and use the dataset.

Using the Python Script

To use the example Python script, simply run:

python example_usage.py

Additional Scripts

cleaning_Data.py: Cleans raw text data. Convert to JSON file.py: Converts text files to JSON format. Overall Data Statistics.py: Prints overall statistics of the dataset. Print duplicate alignments.py: Identifies and prints redundant alignments. Remove duplicate alignments from data set.py: Removes duplicate alignments.

Contact

For access to the full 270,000 sentence pairs or any other inquiries, please contact Syed Matla ul Qamar at [email protected].