Arabic Text Diacritization

This repository contains the dataset, helpers, and systems comparison for our paper on Arabic Text Diacritization:

"Arabic Text Diacritization Using Deep Neural Networks", Ali Fadel, Ibraheem Tuffaha, Bara' Al-Jawarneh, and Mahmoud Al-Ayyoub, ICCAIS 2019.

Files

dataset

• train.txt - Contains 50,000 lines of diacritized Arabic text which can be used as training dataset
• val.txt - Contains 2,500 lines of diacritized Arabic text which can be used as validation dataset
• test.txt - Contains 2,500 lines of diacritized Arabic text which can be used as testing dataset

helpers

• constants
  • ARABIC_LETTERS_LIST.pickle - Contains list of Arbaic letters
  • CLASSES_LIST.pickle - Contains list of all possible classes
  • DIACRITICS_LIST.pickle - Contains list of all diacritics
• count_characters.py - Counts the number of Arabic letters and diacritics in a file
• count_fathatan.py - Counts the number of fathatan occurrences before and after Alif in all files from a folder
• diacritization_stat.py - Calculates DER and WER using the gold data and the predicted output
• diacritics_rate_extractor.py - Keeps lines with p% diacritics to Arabic characters rate or more in all files from a folder
• file_lookup.py - Searches for a line in all files from a folder
• fix_fathatan.py - Changes after-Alif fathatan to before-Alit fathatan in a file
• remove_diacritics.py - Removes diacritics from a file
• transliteration.py - Converts a file from Arabic text to Buckwalter transliteration and vice-versa
• pre_process_tashkeela_corpus.ipynb - Pre-process Tashkeela Corpus data

existing_systems

• ali-soft - Contains some bugs that exist in Ali-Soft system
• farasa - Contains Farasa system output, fixed output, and DER/WER statistics
• harakat - Contains Harakat system testing script, output, fixed output, and DER/WER statistics
• madamira - Contains MADAMIRA system output, fixed output, and DER/WER statistics
• mishkal - Contains Mishkal system output, fixed output, and DER/WER statistics
• shakkala - Contains Shakkala system data splitting script, output, fixed output, and DER/WER statistics
• tashkeela_model - Contains Tashkeela-Model system output, fixed output, and DER/WER statistics for each n-gram model provided by them

Note: All codes in this repository tested on Ubuntu 18.04

Contributors

1. Ali Hamdi Ali Fadel.
   
2. Ibraheem Tuffaha.
   
3. Bara' Al-Jawarneh.
   
4. Mahmoud Al-Ayyoub.
   

License

The project is available as open source under the terms of the MIT License.