--- license: apache-2.0 dataset_info: - config_name: default features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 237943411 num_examples: 991371 download_size: 119013237 dataset_size: 237943411 - config_name: KisatiAna features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 1785984 num_examples: 2975 download_size: 901051 dataset_size: 1785984 - config_name: TN_RAP_lyrics features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 366101 num_examples: 2590 download_size: 189994 dataset_size: 366101 - config_name: TunHistoires features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 119915 num_examples: 111 download_size: 61071 dataset_size: 119915 - config_name: Chakhabitt features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 178248 num_examples: 249 download_size: 83881 dataset_size: 178248 - config_name: TSAC features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 1503515 num_examples: 8416 download_size: 815371 dataset_size: 1503515 - config_name: TunSwitchTunisiaOnly features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 89415494 num_examples: 380546 download_size: 47525584 dataset_size: 89415494 - config_name: HkayetTounsia features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 7975055 num_examples: 23583 download_size: 3982070 dataset_size: 7975055 - config_name: Tweet_TN features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 7292912 num_examples: 39637 download_size: 3097349 dataset_size: 7292912 - config_name: T-HSAB features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 1714013 num_examples: 6024 download_size: 942066 dataset_size: 1714013 - config_name: Tunisien_Dialect_Summary-llama2 features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 174968 num_examples: 95 download_size: 67265 dataset_size: 174968 - config_name: TRCD features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 88638 num_examples: 145 download_size: 41440 dataset_size: 88638 - config_name: CombinedNames features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 12642788 num_examples: 45345 download_size: 2250853 dataset_size: 12642788 - config_name: TuDiCOI features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 310526 num_examples: 3426 download_size: 117595 dataset_size: 310526 - config_name: TunBERT features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 13815487 num_examples: 67186 download_size: 7341958 dataset_size: 13815487 - config_name: Derja_tunsi features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 880033 num_examples: 13037 download_size: 356413 dataset_size: 880033 - config_name: TunSwitchCodeSwitching features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 96432435 num_examples: 394160 download_size: 49638704 dataset_size: 96432435 - config_name: Lbachch features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 212542 num_examples: 900 download_size: 108064 dataset_size: 212542 - config_name: HkayetErwi features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 787932 num_examples: 946 download_size: 310033 dataset_size: 787932 - config_name: BRAD features: - name: ids dtype: string - name: text dtype: string - name: text_raw dtype: string splits: - name: train num_bytes: 2246825 num_examples: 2000 download_size: 1182475 dataset_size: 2246825 configs: - config_name: default data_files: - split: train path: data/*/train-* - config_name: HkayetErwi data_files: - split: train path: data/HkayetErwi/train-* - config_name: TunSwitchCodeSwitching data_files: - split: train path: data/TunSwitchCodeSwitching/train-* - config_name: Derja_tunsi data_files: - split: train path: data/Derja_tunsi/train-* - config_name: TunBERT data_files: - split: train path: data/TunBERT/train-* - config_name: TuDiCOI data_files: - split: train path: data/TuDiCOI/train-* - config_name: CombinedNames data_files: - split: train path: data/CombinedNames/train-* - config_name: TRCD data_files: - split: train path: data/TRCD/train-* - config_name: Tunisien_Dialect_Summary-llama2 data_files: - split: train path: data/Tunisien_Dialect_Summary-llama2/train-* - config_name: KisatiAna data_files: - split: train path: data/KisatiAna/train-* - config_name: TN_RAP_lyrics data_files: - split: train path: data/TN_RAP_lyrics/train-* - config_name: TunHistoires data_files: - split: train path: data/TunHistoires/train-* - config_name: Chakhabitt data_files: - split: train path: data/Chakhabitt/train-* - config_name: TSAC data_files: - split: train path: data/TSAC/train-* - config_name: TunSwitchTunisiaOnly data_files: - split: train path: data/TunSwitchTunisiaOnly/train-* - config_name: HkayetTounsia data_files: - split: train path: data/HkayetTounsia/train-* - config_name: Tweet_TN data_files: - split: train path: data/Tweet_TN/train-* - config_name: T-HSAB data_files: - split: train path: data/T-HSAB/train-* - config_name: Lbachch data_files: - split: train path: data/Lbachch/train-* - config_name: BRAD data_files: - split: train path: data/BRAD/train-* --- LinTO DataSet Text for Arabic Tunisian A collection of Tunisian dialect text for STT task. This is the first packaged version of the text datasets used to train the Linto Tunisian dialect with code-switching STT (linagora/linto-asr-ar-tn-0.1). * [Dataset Summary](#dataset-summary) * [Dataset composition](#dataset-composition) * [Sources](#sources) * [Data Table](#data_table) * [Data sources](#data-sources) * [Content Types](#content-types) * [Languages and Dialects](#languages-and-dialects) * [Example use (python)](#example-use-python) * [License](#license) * [Citations](#citations) ## Dataset Summary The **LinTO DataSet Text for Arabic Tunisian** is a diverse collection of text sources, including music (e.g., RAP lyrics), documentaries, names, stories, and more. This dataset was curated to support the development of a language model for the Tunisian Arabic ASR system. ## Dataset Composition The **LinTO DataSet Audio for Arabic Tunisian v0.1** comprises a diverse range of text content collected from multiple sources. Below is a breakdown of the dataset’s composition: ### Sources - **Hugging Face Datasets**: Various datasets obtained from the Hugging Face platform. - **GitHub**: Various datasets obtained from the GitHub. - **Websites**: Text collected from various online sources, including film script websites and story-sharing platforms. ### Data Table | **subset** | **Lines** | **words** | **#Unique Words** | | --- | --- | --- | --- | | [Chakhabit](https://chakhabitt.blogspot.com/) | 387 | 42481 | 13949 | | [CombinedNames](#) | 45345 | 490900 | 10936 | | [HkayetErwi](https://hikayattunisien.blogspot.coml) | 966 | 43957 | 8045 | | [HkayetTounsia](https://chroniquechroniyet.blogspot.com/) | 27365 | 436848 | 68975 | | [Lbachch](https://lbachch.blogspot.com/) | 912 | 10376 | 5072 | | [Derja_tunsi](https://www.kaggle.com/datasets/khawlajlassi/drejja-to-english?resource=download) | 13037 | 19963 | 11324 | | [T-HSAB](https://github.com/Hala-Mulki/T-HSAB-A-Tunisian-Hate-Speech-and-Abusive-Dataset) | 6024 | 70736 | 27171 | | [TN_RAP_lyrics](https://www.lyricstn.tn/search/label/RAP?m=1) | 2590 | 15679 | 7441 | | [TRCD](https://arxiv.org/abs/1704.00051) | 145 | 4036 | 1575 | | [TSAC](https://github.com/fbougares/TSAC) | 8479 | 69606 | 19457 | | [TuDiCOI](https://huggingface.co/datasets/arbml/TuDiCoI) | 3426 | 10570 | 1392 | | [TunBERT](https://github.com/instadeepai/tunbert) | 67219 | 579748 | 117756 | | [TunHistoires](https://tunhistoires.blogspot.com/) | 857 | 11433 | 4649 | | [TunSwitchCodeSwitching](https://zenodo.org/records/8342762) | 394163 | 4406672 | 192530 | | [TunSwitchTunisiaOnly](https://zenodo.org/records/8342762) | 380546 | 4211374 | 178108 | | [Tunisien_Dialect_Summary-llama2](https://huggingface.co/datasets/medmabfc/Tunisien_Dialect_Summary-llama2-test26) | 95 | 8962 | 2814 | | Tweet_TN | 39637 | 329479 | 37275 | | [KisatiAna](https://kisatiana.blogspot.com/) | 3024 | 95504 | 22986 | | [BRAD](https://github.com/4mekki4/arabic-nlp-da/blob/main/data) | 2000 | 109738 | 31976 | ### Data Proccessing: - **Word Normalization**: Standardizes spelling variations and normalizes different forms of the same word to a consistent format. This is essential for improving model performance by reducing vocabulary size and minimizing ambiguity in the dataset. - **Number Conversion**: Converts numbers to their textual representations (e.g., "123" to "one hundred twenty-three") to maintain consistency in the dataset, particularly in spoken text applications where numeric characters are read as words. - **Punctuation Removal**: Strips out punctuation marks that are not essential for understanding, such as commas and periods, which helps focus the model on the core content of the text. ### Content Types - **Music**: Includes RAP lyrics. - **Sentiment Analysis**: Text of sentiment analysis - **Documentaries**: Text from documentaries about history and nature. - **Tweets**: Tweets sourced from X (Twitter). - **Lessons**: Learning resources for the Tunisian dialect. - **Others**: Mixed text with various subjects. ### Languages and Dialects - **Tunisian Arabic**: The primary focus of the dataset, including Tunisian Arabic and some Modern Standard Arabic (MSA). - **French**: Some instances of French code-switching. - **English**: Some instances of English code-switching. ## Example use (python) - **Load the dataset in python**: ```python from datasets import load_dataset # dataset will be loaded as a DatasetDict of train and test dataset = load_dataset("linagora/textual-ar-tn-dataset") ``` Check the containt of dataset: ```python example = dataset['train'][1] text = example['text'] print(f"text: {text}") ``` **Example** ```bash text: اكبر عدو للانسان حب الذات وخاصه إذا وصل حد ان تعمى البصيره اكبر مرض حينما تحب ذاتك وتدور فقط فى فلكك وبلا وعى ترجع كل مايدور حولك الى هل سيعجبك ام لا ولكن هناك من احبهم الله ورضى عنهم لشي لا يعلمه الا هو فينير بصيرتهم الى من حولهم ويذهر انفسهم من مرض الايجو الخطوات و الطريق صعب وهنا جهاد النفس و تربيه الذات وتحمل مشقه العلاج و الصبر عليه ``` ## License Given that some of the corpora used for training and evaluation are available only under Apache 2 licenses. ## Citations When using the **LinTO DataSet Audio for Arabic Tunisian v0.1** corpus, please cite this page: ```bibtex @misc{linagora2024Linto-tn, author = {Hedi Naouara and Jérôme Louradour and Jean-Pierre Lorré and Sarah zribi and Wajdi Ghezaiel}, title = {LinTO DataSet Text for Arabic Tunisian}, year = {2024}, publisher = {HuggingFace}, journal = {HuggingFace}, howpublished = {\url{https://huggingface.co/datasets/linagora/linto-dataset-text-ar-tn}}, } ``` ```bibtex @misc{abdallah2023leveraging, title={Leveraging Data Collection and Unsupervised Learning for Code-switched Tunisian Arabic Automatic Speech Recognition}, author={Ahmed Amine Ben Abdallah and Ata Kabboudi and Amir Kanoun and Salah Zaiem}, year={2023}, eprint={2309.11327}, archivePrefix={arXiv}, primaryClass={eess.AS} } ``` ```bibtex @article{DBLP:journals/corr/ChenFWB17, author = {Danqi Chen and Adam Fisch and Jason Weston and Antoine Bordes}, title = {Reading Wikipedia to Answer Open-Domain Questions}, journal = {CoRR}, volume = {abs/1704.00051}, year = {2017}, url = {http://arxiv.org/abs/1704.00051}, eprinttype = {arXiv}, eprint = {1704.00051}, timestamp = {Thu, 14 Dec 2023 18:03:42 +0100}, biburl = {https://dblp.org/rec/journals/corr/ChenFWB17.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} } ```