---
pretty_name: 'Dialect of Tunisia a Collection of Works'
---
# Dialect of Tunisia - a Collection of Works :


This repository compiles a variety of Tunisian NLP projects to create a comprehensive text-based dataset for diverse applications. It exclusively features Tunisian dialect written in Arabic script and includes code-switching with other languages such as French, English, and Italian. The corpus comprises a total of 70 million GPT-4 tokens. Light Preprocessing has been done on these source datasets to remove negative text , repetitive tokens , emojies, and non-arabic letters dialect


### Data :
Below are the individual datasets that are combined in this work 

* ###  [Arbi-Houssem/Tunisian_dataset_STT-TTS15s_filtred_organiser_Mixed](https://huggingface.co/datasets/Arbi-Houssem/Tunisian_dataset_STT-TTS15s_filtred_organiser_Mixed) : 
A speech to text dataset where we consider the target sentences ,we keep transcript as  is 

* ### [linagora/linto-asr-ar-tn-0.1](https://huggingface.co/linagora/linto-asr-ar-tn-0.1) :
An ASR dataset representing a collection of various sources. Sources are . Youtube_TNScrapped, TunswitchTO , TunswitchCS, ApprendreLeTunisien ,Taric, OneStorym we keep sentences as is

 * ### [Tunisian Arabic Dialects Identification(TADI): ](https://github.com/instadeepai/tunbert) :
It is a binary classification task consisting of classifying Tunisian dialect and Non Tunisian dialect from an Arabic dialectical dataset.
we only extract rows identified as tunisian dialect
 * ### Tunisian Algerian Dialect(TAD) (https://github.com/instadeepai/tunbert): 
It is a binary classification task consisting of classifying Tunisian dialect and Algerian dialect from an Arabic dialectical dataset.
we only extract rows identified as tunisian dialect

 * ###  [khaled123/tunninjaar](https://huggingface.co/datasets/khaled123/tunninjaar):
A dataset extracted from : https://derja.ninja/

 * ###  [Hala-Mulki T-HSAB-A-Tunisian-Hate-Speech-and-Abusive-Dataset](https://github.com/Hala-Mulki/T-HSAB-A-Tunisian-Hate-Speech-and-Abusive-Dataset):
the first Arabic Tunisian Hate Speech and Abusive Language Dataset proposed in the The 7th International Conference on Arabic Language Processing October
only comsider rows marked as non hate speech

 * ### [Tunisian Reading Comprehension Dataset](https://github.com/instadeepai/tunbert):
A Question-Answering dataset for Tunisian dialect. Use of dialectal version of the Tunisian constitution . It is composed of 144 documents where each document has exactly 3 paragraphs and three Question-Answer pairs are assigned to each paragraph. 

 * ### [Naim Mhedhbi Tunisian Dialect Corpus v0](https://www.kaggle.com/datasets/naim99/ts-naim-mhedhbi) :
40000 comments and posts from facebook, labeled  as  positive / negative/ neutral , 
we only consider positive and neutral rows

 * ### TSAC (Tunisian Sentiment Analysis Corpus)(https://paperswithcode.com/dataset/tsac)
Tunisian Sentiment Analysis Corpus (TSAC) is a Tunisian Dialect corpus of 17.000 comments from Facebook.

 * ###  [khaled123/Testtun](https://huggingface.co/datasets/khaled123/Testtun):

 * ###  [khaled123/Tuniset](https://huggingface.co/datasets/khaled123/tuniset):

| Dataset  | Token Count |
|----------|----------|
| arbi_houssem-tunisian-dataset_stt_tts15s_filtered1.0 1    | 129,207  | 
| maal-32173-Tunisian_Revolution_Arabic-2011-2022    | 11,538,264  | 
| TAD   | 41,434   | 
| TADI   | 891,750   | 
| testtun-khaldoun   | 18,356,271   | 
| thsab   | 289,310   | 
|  ts-naim-mhedhbi   | 224,434   | 
| tsac   | 97,085   | 
| tuninjar   | 13,056,137   | 
| tuniset   | 22,584,964   | 
| linagoraSTT   | 2,003,450   | 
| Total   | 69,422,659   | 

Count of Unique tokens = 534,678