gsarti/clean_mc4_it

Dataset Card for Clean Italian mC4 🇮🇹

Dataset Summary

A thoroughly cleaned version of the Italian split of the multilingual colossal, cleaned version of Common Crawl's web crawl corpus (mC4). Based on the Common Crawl dataset. The original version was prepared by AllenAI, hosted at the address https://huggingface.co/datasets/allenai/c4, with subsequent preprocessing performed by Gabriele Sarti following a standard procedure for all dataset shards.

Preprocessing

The preprocessing of the dataset follows the procedure used by Yeb Havinga for training the model t5-base-dutch on a portion of the cleaned Dutch split of mC4. The original code, that was adapted for Italian in this case, is available on GitLab. In summary, the preprocessing procedure includes:

• Removing documents containing words from a selection of the Italian and English List of Dirty Naught Obscene and Otherwise Bad Words.

• Removing sentences containing:

  • Less than 3 words.

  • A word longer than 1000 characters.

  • An end symbol not matching end-of-sentence punctuation.

  • Strings associated to javascript code (e.g. {), lorem ipsum, policy information in Italian or English.

• Removing documents (after sentence filtering):

  • Containing less than 5 sentences.

  • Containing less than 500 or more than 50'000 characters.

  • Not identified as prevalently Italian by the LangDetect package.

Using parallel processing with 96 CPU cores on a TPUv3 via Google Cloud to perform the complete clean of all the original Italian shards of mC4 (1024 of ~220Mb train, 8 of ~24Mb validation) required roughly 10 hours due to the demanding steps of sentence tokenization and language detection. The total size of compressed .json.gz files is roughly halved after the procedure.

Dataset Structure

Data Instances

An example from the dataset:

{
  'timestamp': '2020-02-22T22:24:31Z', 
  'url': 'https://altreconomia.it/una-rotonda-sul-pane/', 
  'text': 'Per raggiungere il campo attraversiamo la striscia d’asfalto che porta verso la provinciale numero 13. Mettiamo a rischio la nostra incolumità in un territorio di auto e camion. Sullo sfondo, i profili della Grigna e del Resegone. Più vicini, quelli del solito ipermercato di provincia, e delle villette a schiera che avanzano tra le coltivazioni. È lo sprawling, l’avanzata del cemento.\\nDa questo lato dalla strada, invece, è ancora regno contadino. Almeno per ora. Torniamo a Caponago (Mb), Brianza pura, dove ha avuto i natali il progetto “Spiga e madia”. Ne parlammo su Ae nel gennaio 2009: in un territorio “spaesato”, il Comitato “verso il Distretto di economia solidale della Brianza” (Desbri) e la “Retina” dei gruppi di acquisto locali danno vita a un progetto di produzione di frumento, molitura, panificazione e distribuzione in un raggio di 20 chilometri. Si comincia da zero, nel 2007, senza alcun di finanziamento, quando una famiglia del [...]. Il giochino vale almeno 3 miliardi di euro all’anno. La misura, introdotta in via straordinaria con la finanziaria 2005, è stata prorogata anche con l’ultimo decreto “milleproroghe”.'
}

Data Fields

The data contains the following fields:

• url: url of the source as a string
• text: text content as a string
• timestamp: timestamp of extraction as a string

Data Splits

To build mC4, the original authors used CLD3 to identify over 100 languages. For Italian, the whole corpus of scraped text was divided in 1032 jsonl files, 1024 for training following the naming style c4-it.tfrecord-0XXXX-of-01024.json.gz and 8 for validation following the naming style c4-it-validation.tfrecord-0000X-of-00008.json.gz. The full set of preprocessed files takes roughly 215GB of disk space to download with Git LFS.

For ease of use under different storage capacities, the following incremental splits are available (sizes are estimates). Important: The sizes in GB represent the estimated weight for :

split	train size (docs, words, download + preproc disk space)	validation size
tiny	10M docs, 4B words (9 GB + 27 GB)	12k docs
small	20M docs, 8B words (18 GB + 54 GB)	24k docs
medium	50M docs, 20B words (47 GB + 135 GB)	48k docs
large	75M docs, 30B words (71 GB + 203 GB)	72k docs
full	103M docs, 41B words (109 GB + 279 GB)	96k docs

You can load any subset like this:

from datasets import load_dataset

mc4_it_tiny = load_dataset("gsarti/clean_mc4_it", "tiny")

Since splits are quite large, you may want to traverse them using the streaming mode available starting from 🤗 Datasets v1.9.0:

from datasets import load_dataset

mc4_it_full_stream = load_dataset("gsarti/clean_mc4_it", "full", split='train', streaming=True)
print(next(iter(mc4_it_full_stream))) # Prints the example presented above

Dataset Creation

Refer to the original paper for more considerations regarding the choice of sources and the scraping process for creating mC4.

Considerations for Using the Data

Social Impact of Dataset

With more than 200GB of cleaned Italian text and more than 41B estimated words, this is by far the largest available corpus for the Italian language. The second largest dataset available is OSCAR, which is only 69GB in size for its deduplicated variant. Using this corpus for training language models with adequate computational resources will allow researchers to reach parity with the performances observed for the English language. This can in turn have important repercussions for the development of commercial language technology applications for the Italian language.

Discussion of Biases

Despit the cleaning procedure aimed at removing vulgarity and profanity, it must be considered that model trained on this scraped corpus will inevitably reflect biases present in blog articles and comments on the Internet. This makes the corpus especially interesting in the context of studying data biases and how to limit their impacts.

Additional Information

Dataset Curators

Authors at AllenAI are the original curators for the mc4 corpus. For inquiries or requests regarding the Italian cleaned portion contained in this repository, please contact me at gabriele.sarti996@gmail.com

Licensing Information

AllenAI are releasing this dataset under the terms of ODC-BY. By using this, you are also bound by the Common Crawl terms of use in respect of the content contained in the dataset.

Citation Information

If you use this dataset in your work, please cite us and the original mC4 authors as:

@inproceedings{sarti-nissim-2024-it5-text,
    title = "{IT}5: Text-to-text Pretraining for {I}talian Language Understanding and Generation",
    author = "Sarti, Gabriele  and
      Nissim, Malvina",
    editor = "Calzolari, Nicoletta  and
      Kan, Min-Yen  and
      Hoste, Veronique  and
      Lenci, Alessandro  and
      Sakti, Sakriani  and
      Xue, Nianwen",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italy",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.823",
    pages = "9422--9433",
    abstract = "We introduce IT5, the first family of encoder-decoder transformer models pretrained specifically on Italian. We document and perform a thorough cleaning procedure for a large Italian corpus and use it to pretrain four IT5 model sizes. We then introduce the ItaGen benchmark, which includes a broad range of natural language understanding and generation tasks for Italian, and use it to evaluate the performance of IT5 models and multilingual baselines. We find monolingual IT5 models to provide the best scale-to-performance ratio across tested models, consistently outperforming their multilingual counterparts and setting a new state-of-the-art for Italian language generation.",
}

@inproceedings{xue-etal-2021-mt5,
    title = "m{T}5: A Massively Multilingual Pre-trained Text-to-Text Transformer",
    author = "Xue, Linting  and
      Constant, Noah  and
      Roberts, Adam  and
      Kale, Mihir  and
      Al-Rfou, Rami  and
      Siddhant, Aditya  and
      Barua, Aditya  and
      Raffel, Colin",
    booktitle = "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
    month = jun,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.naacl-main.41",
    doi = "10.18653/v1/2021.naacl-main.41",
    pages = "483--498",
}