Tahdheeb 0.1.1

An Arabic Preprocessing Toolkit for Pretraining of LLM

Tahdheeb (تهذیب)

Overview

Tahdheeb (تهذیب) is an Arabic Preprocessing Library developed at Qatar Computing Research. This library provides tools for preprocessing Arabic text, including normalization, diacritics handling, number normalization, punctuation correction, and more.

Develpers

• Ehsaneddin Asgari
• Yassine Elkheir

Features

• Normalization: Standardizes Arabic text by normalizing character variations and numbers.
• Diacritics Handling: Removes or processes Arabic diacritics.
• Punctuation Correction: Corrects punctuation usage in Arabic text.
• Determine Arabic Ratio: Output the arabic ration.

Installation

You can install Tahdheeb using pip:

pip install src

Example Usage

The following example demonstrates how to use the Tahdheeb library for preprocessing Arabic text, including fixing punctuation issues, improving text structure, and normalizing the text.

Code Example

from src.preprocessing.text_content import ArabicTextProcess
from src.preprocessing.text_verification import ArabicTextVerify
from src.preprocessing.text_structure import ArabicStructure

text = 'الـــــــــسلام    عليكم . كيـــف    حالك ؟ أتـــمنى أن تكون بــــــخير .أ نحنو  هنا للمســـــاعدة ...هل  تريــــد شيء مـــا؟  الرجـــــــاء    إبلاغن ا بــــأي وقــــت! في ٢دقيقه شكرا ، جزيــــلا Thanks!'

# Initialize the classes
ATB = ArabicStructure()
ASP = ArabicTextProcess()
ATP = ArabicTextVerify()

# Check the ratio of Arabic text
print('Ratio of Arabic text:', ATP.check_arabic_ratio(text))

print('\nText before structure improvement:\n', text, '\n-----------\n')

# Preprocess the text to fix structure and remove English text
text = ATB.preprocess(text, remove_english=True)
print('\nText after structure improvement:\n', text, '\n-----------\n')

# Normalize the text (basic normalization)
text = ASP.get_arabic_normal(text, extensive_normalization=False)
print('\nText after normalizing:\n', text, '\n-----------\n')

# Normalize the text (extensive normalization)
text = ASP.get_arabic_normal(text, extensive_normalization=True)
print('\nText after extensive normalizing:\n', text, '\n-----------\n')

text = 'بِسْمِ اللَّهِ الرَّحْمَٰنِ الرَّحِيمِ'

print('\ntext before diac removal:\n', text, '\n-----------\n')
text = ASP.get_arabic_normal(text,  extensive_normalization=False, diacritics_removal=True)
print('\ntext after diac removal:\n', text, '\n-----------\n')

Output

Ratio of Arabic text: 0.7107843137254902

text before struct improving:
 الـــــــــسلام    عليكم . كيـــف    حالك ؟ أتـــمنى أن تكون بــــــخير .أ نحنو  هنا للمســـــاعدة ...هل  تريــــد شيء مـــا؟  الرجـــــــاء    إبلاغن ا بــــأي وقــــت! في ٢دقيقه شكرا ، جزيــــلا Thanks! 
-----------


text after struct improving:
 الـــــــــسلام عليكم. كيـــف حالك؟ أتـــمنى أن تكون بــــــخير. أ نحنو هنا للمســـــاعدة... هل تريــــد شيء مـــا؟ الرجـــــــاء إبلاغن ا بــــأي وقــــت! في ٢ دقيقه شكرا، جزيــــلا! 
-----------


text after normalizing:
 السلام عليكم. كيف حالك؟ أتمنى أن تكون بخير. أ نحنو هنا للمساعدة... هل تريد شيء ما؟ الرجاء إبلاغن ا بأي وقت! في 2 دقيقه شكرا، جزيلا! 
-----------


text after extensive normalizing:
 السلام علىكم. كىف حالك؟ اتمنى ان تكون بخىر. ا نحنو هنا للمساعده... هل ترىد شىء ما؟ الرجاء ابلاغن ا باى وقت! فى 2 دقىقه شكرا، جزىلا! 
-----------

text before diac removal:
 بِسْمِ اللَّهِ الرَّحْمَٰنِ الرَّحِيمِ 
-----------


text after diac removal:
 بسم الله الرحمن الرحيم 
-----------

Explanation

Import Libraries

from src.preprocessing.text_content import ArabicTextProcess
from src.preprocessing.text_verification import ArabicTextVerify
from src.preprocessing.text_structure import ArabicStructure

Original Text

The sample text contains various errors such as character repetition, incorrect punctuation, and mixed language content.

text = 'الـــــــــسلام    عليكم . كيـــف    حالك ؟ أتـــمنى أن تكون بــــــخير .أ نحنو  هنا للمســـــاعدة ...هل  تريــــد شيء مـــا؟  الرجـــــــاء    إبلاغن ا بــــأي وقــــت! في ٢دقيقه شكرا ، جزيــــلا Thanks!'

Initialize the Classes

Initialize the classes for structure improvement, text processing, and text verification.

ATB = ArabicStructure()
ASP = ArabicTextProcess()
ATP = ArabicTextVerify()

Check the Ratio of Arabic Text

Verify how much of the text is in Arabic.

print('Ratio of Arabic text:', ATP.check_arabic_ratio(text))

Before Structure Improvement

Print the original text before any processing.

print('\nText before structure improvement:\n', text, '\n-----------\n')

Structure Improvement

Preprocess the text to fix punctuation issues and remove English text.

text = ATB.preprocess(text, remove_english=True)
print('\nText after structure improvement:\n', text, '\n-----------\n')

Basic Normalization

Normalize the text with basic normalization.

text = ASP.get_arabic_normal(text, extensive_normalization=False)
print('\nText after normalizing:\n', text, '\n-----------\n')

Extensive Normalization

Normalize the text with extensive normalization.

text = ASP.get_arabic_normal(text, extensive_normalization=True)
print('\nText after extensive normalizing:\n', text, '\n-----------\n')

Output

This example will output the following information at each step:

1. The ratio of Arabic text in the original input.
2. The original text before any processing.
3. The text after fixing structural issues and removing English text.
4. The text after basic normalization.
5. The text after extensive normalization.

This step-by-step process ensures that the Arabic text is cleaned, normalized, and ready for further processing or analysis.

Notes

Diacretics

Unicode	Diacritic	Name (English)	Name (Arabic)	Description
\u064b	ً	Tanwin Fathah	تَنوين الفتح	Indicates a nun sound after the vowel (an).
\u064c	ٌ	Tanwin Dammah	تَنوين الضم	Indicates a nun sound after the vowel (un).
\u064d	ٍ	Tanwin Kasrah	تَنوين الكسر	Indicates a nun sound after the vowel (in).
\u064e	َ	Fathah	فتحة	Represents a short "a" sound.
\u064f	ُ	Dammah	ضمة	Represents a short "u" sound.
\u0650	ِ	Kasrah	كسرة	Represents a short "i" sound.
\u0651	ّ	Shaddah	شدة	Indicates gemination or doubling of a consonant.
\u0652	ْ	Sukun	سكون	Indicates the absence of a vowel.
\u0653	ٓ	Maddah	مدّة	Indicates a long vowel sound.
\u0670	ٰ	Superscript Alif	ألف خنجرية	Indicates a short "a" sound, mainly in Quranic text.