README.md

metadata

license: cc-by-nc-3.0
task_categories:
  - text-classification
  - token-classification
  - translation
language:
  - ar
tags:
  - NLP
  - Tunisia
  - Arabic
pretty_name: MADAR-TUN LINGUISTIC ANNOTATIONS
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: index
      dtype: int64
    - name: arabish
      dtype: string
    - name: class
      dtype: string
    - name: words
      dtype: string
    - name: tokens
      dtype: string
    - name: pos
      dtype: string
    - name: lem
      dtype: string
  splits:
    - name: train
      num_bytes: 1990921
      num_examples: 27123
    - name: test
      num_bytes: 221237
      num_examples: 3014
  download_size: 607675
  dataset_size: 2212158

MADAR-TUN LINGUISTIC ANNOTATIONS

Note: This dataset is a direct mirror of the original MADAR-TUN repository created by Eli Gugliotta and collaborators.
It is hosted here on Hugging Face for easier access and loading through the datasets library.
All credit for the dataset, annotations, and methodology goes to the original authors.
The dataset is redistributed under the same license as the original repository.

---

This repository provides a brief description and houses the linguistic annotations for the Tunisian part of the MADAR corpus (Bouamor et al., 2018). These annotations were created during a research project outlined in the following paper, which has been accepted for presentation at the LDK conference scheduled to be held in Vienna in September 2023:

• Gugliotta, E. and Dinarelli, M. (2023). An Empirical Analysis of Task Relations in the Multi-Task Annotation of an Arabizi Corpus

The annotations presented here are the outcome of the same procedure used for the TArC corpus. For additional information and in-depth details, please refer to the TArC repository, or consult the following papers. The papers shed light on the methodology employed and provide a comprehensive understanding of the annotation process carried out for both the MADAR and TArC corpora.

• Gugliotta, E. and Dinarelli, M. (2022). Tarc: Tunisian arabish corpus first complete release. LREC2022.

• Gugliotta, E. et al., (2020). Multi-Task Sequence Prediction For Tunisian Arabizi Multi-Level Annotation.

Overview of DATA

In this repository, you will find a collection of linguistic annotations, generated semi-automatically using a Multi-Task Sequence Prediction System and manually validated:

• Transliteration into Arabizi (Arabish level)
• Token classification: Tokens are categorized into three groups - arabizi (representing Arabic words written in Latin script), foreign (indicating foreign words), and emotag (denoting emoticons or smileys tags).
• Normalization in Arabic Script of the tokens classified as arabizi (following the CODA* convention (Habash et al., 2018)).
• Tokenization of the CODAfied texts.
• Part-of-Speech tagging for the arabizi tokens.
• Lemmatization, in CODA* (Arabic-script)

MADAR-TUN-annotation numbers:

SENTENCES		LEMMA
3,990		3,033

Citation

Please cite this work as:

@inproceedings{gugliotta-etal-wanlp2020, 
    title={An Empirical Analysis of Task Relations in the Multi-Task Annotation of an Arabizi Corpus}, 
    author={Gugliotta, Elisa and Dinarelli, Marco}, 
    booktitle={The 4th Conference on Language, Data and Knowledge (LDK 2023)}, 
    year={2023},
}