{"Sakonii--nepali_LM_dataset_wiki_cc100_OSCAR_combined": {"description": "This corpus is an attempt to recreate the dataset used for training XLM-R. This corpus comprises of monolingual data for 100+ languages and also includes data for romanized languages (indicated by *_rom). This was constructed using the urls and paragraph indices provided by the CC-Net repository by processing January-December 2018 Commoncrawl snapshots. Each file comprises of documents separated by double-newlines and paragraphs within the same document separated by a newline. The data is generated using the open source CC-Net repository. No claims of intellectual property are made on the work of preparation of the corpus.\n\n\nThe Open Super-large Crawled ALMAnaCH coRpus is a huge multilingual corpus obtained by language classification and filtering of the Common Crawl corpus using the goclassy architecture.\n\n", "citation": "@inproceedings{conneau-etal-2020-unsupervised,\n    title = \"Unsupervised Cross-lingual Representation Learning at Scale\",\n    author = \"Conneau, Alexis  and\n      Khandelwal, Kartikay  and\n      Goyal, Naman  and\n      Chaudhary, Vishrav  and\n      Wenzek, Guillaume  and\n      Guzm{'a}n, Francisco  and\n      Grave, Edouard  and\n      Ott, Myle  and\n      Zettlemoyer, Luke  and\n      Stoyanov, Veselin\",\n    booktitle = \"Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics\",\n    month = jul,\n    year = \"2020\",\n    address = \"Online\",\n    publisher = \"Association for Computational Linguistics\",\n    url = \"https://www.aclweb.org/anthology/2020.acl-main.747\",\n    doi = \"10.18653/v1/2020.acl-main.747\",\n    pages = \"8440--8451\",\n    abstract = \"This paper shows that pretraining multilingual language models at scale leads to significant performance gains for a wide range of cross-lingual transfer tasks. We train a Transformer-based masked language model on one hundred languages, using more than two terabytes of filtered CommonCrawl data. Our model, dubbed XLM-R, significantly outperforms multilingual BERT (mBERT) on a variety of cross-lingual benchmarks, including +14.6{%} average accuracy on XNLI, +13{%} average F1 score on MLQA, and +2.4{%} F1 score on NER. XLM-R performs particularly well on low-resource languages, improving 15.7{%} in XNLI accuracy for Swahili and 11.4{%} for Urdu over previous XLM models. We also present a detailed empirical analysis of the key factors that are required to achieve these gains, including the trade-offs between (1) positive transfer and capacity dilution and (2) the performance of high and low resource languages at scale. Finally, we show, for the first time, the possibility of multilingual modeling without sacrificing per-language performance; XLM-R is very competitive with strong monolingual models on the GLUE and XNLI benchmarks. We will make our code and models publicly available.\",\n}\n@inproceedings{wenzek-etal-2020-ccnet,\n    title = \"{CCN}et: Extracting High Quality Monolingual Datasets from Web Crawl Data\",\n    author = \"Wenzek, Guillaume  and\n      Lachaux, Marie-Anne  and\n      Conneau, Alexis  and\n      Chaudhary, Vishrav  and\n      Guzm{'a}n, Francisco  and\n      Joulin, Armand  and\n      Grave, Edouard\",\n    booktitle = \"Proceedings of the 12th Language Resources and Evaluation Conference\",\n    month = may,\n    year = \"2020\",\n    address = \"Marseille, France\",\n    publisher = \"European Language Resources Association\",\n    url = \"https://www.aclweb.org/anthology/2020.lrec-1.494\",\n    pages = \"4003--4012\",\n    abstract = \"Pre-training text representations have led to significant improvements in many areas of natural language processing. The quality of these models benefits greatly from the size of the pretraining corpora as long as its quality is preserved. In this paper, we describe an automatic pipeline to extract massive high-quality monolingual datasets from Common Crawl for a variety of languages. Our pipeline follows the data processing introduced in fastText (Mikolov et al., 2017; Grave et al., 2018), that deduplicates documents and identifies their language. We augment this pipeline with a filtering step to select documents that are close to high quality corpora like Wikipedia.\",\n    language = \"English\",\n    ISBN = \"979-10-95546-34-4\",\n}\n\n\n@inproceedings{ortiz-suarez-etal-2020-monolingual,\n    title = \"A Monolingual Approach to Contextualized Word Embeddings for Mid-Resource Languages\",\n    author = \"Ortiz Su{'a}rez, Pedro Javier  and\n      Romary, Laurent  and\n      Sagot, Benoit\",\n    booktitle = \"Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics\",\n    month = jul,\n    year = \"2020\",\n    address = \"Online\",\n    publisher = \"Association for Computational Linguistics\",\n    url = \"https://www.aclweb.org/anthology/2020.acl-main.156\",\n    pages = \"1703--1714\",\n    abstract = \"We use the multilingual OSCAR corpus, extracted from Common Crawl via language classification, filtering and cleaning, to train monolingual contextualized word embeddings (ELMo) for five mid-resource languages. We then compare the performance of OSCAR-based and Wikipedia-based ELMo embeddings for these languages on the part-of-speech tagging and parsing tasks. We show that, despite the noise in the Common-Crawl-based OSCAR data, embeddings trained on OSCAR perform much better than monolingual embeddings trained on Wikipedia. They actually equal or improve the current state of the art in tagging and parsing for all five languages. In particular, they also improve over multilingual Wikipedia-based contextual embeddings (multilingual BERT), which almost always constitutes the previous state of the art, thereby showing that the benefit of a larger, more diverse corpus surpasses the cross-lingual benefit of multilingual embedding architectures.\",\n}\n\n@inproceedings{OrtizSuarezSagotRomary2019,\n  author    = {Pedro Javier {Ortiz Su{'a}rez} and Benoit Sagot and Laurent Romary},\n  title     = {Asynchronous pipelines for processing huge corpora on medium to low resource infrastructures},\n  series = {Proceedings of the Workshop on Challenges in the Management of Large Corpora (CMLC-7) 2019. Cardiff, 22nd July 2019},\n  editor    = {Piotr Ba\u0144ski and Adrien Barbaresi and Hanno Biber and Evelyn Breiteneder and Simon Clematide and Marc Kupietz and Harald L{\"u}ngen and Caroline Iliadi},\n  publisher = {Leibniz-Institut f{\"u}r Deutsche Sprache},\n  address   = {Mannheim},\n  doi       = {10.14618/ids-pub-9021},\n  url       = {http://nbn-resolving.de/urn:nbn:de:bsz:mh39-90215},\n  pages     = {9 -- 16},\n  year      = {2019},\n  abstract  = {Common Crawl is a considerably large, heterogeneous multilingual corpus comprised of crawled documents from the internet, surpassing 20TB of data and distributed as a set of more than 50 thousand plain text files where each contains many documents written in a wide variety of languages. Even though each document has a metadata block associated to it, this data lacks any information about the language in which each document is written, making it extremely difficult to use Common Crawl for monolingual applications. We propose a general, highly parallel, multithreaded pipeline to clean and classify Common Crawl by language; we specifically design it so that it runs efficiently on medium to low resource infrastructures where I/O speeds are the main constraint. We develop the pipeline so that it can be easily reapplied to any kind of heterogeneous corpus and so that it can be parameterised to a wide range of infrastructures. We also distribute a 6.3TB version of Common Crawl, filtered, classified by language, shuffled at line level in order to avoid copyright issues, and ready to be used for NLP applications.},\n  language  = {en}\n}\n\n\n", "homepage": "https://data.statmt.org/cc-100/\n\nhttps://oscar-corpus.com\n\n", "license": "\n\n\n    These data are released under this licensing scheme\n    We do not own any of the text from which these data has been extracted.\n    We license the actual packaging of these data under the Creative Commons CC0 license     (\"no rights reserved\") http://creativecommons.org/publicdomain/zero/1.0/\n    To the extent possible under law, Inria has waived all copyright     and related or neighboring rights to OSCAR\n    This work is published from: France.\n\n    Should you consider that our data contains material that is owned by you     and should therefore not be reproduced here, please:\n    * Clearly identify yourself, with detailed contact data such as an address,     telephone number or email address at which you can be contacted.\n    * Clearly identify the copyrighted work claimed to be infringed.\n    * Clearly identify the material that is claimed to be infringing and     information reasonably sufficient to allow us to locate the material.\n\n    We will comply to legitimate requests by removing the affected sources     from the next release of the corpus. ", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": null, "config_name": null, "version": null, "splits": {"train": {"name": "train", "num_bytes": 7133236103.945269, "num_examples": 13141222, "dataset_name": "nepali_LM_dataset_wiki_cc100_OSCAR_combined"}, "test": {"name": "test", "num_bytes": 145576679.05473155, "num_examples": 268189, "dataset_name": "nepali_LM_dataset_wiki_cc100_OSCAR_combined"}}, "download_checksums": null, "download_size": 2971783582, "post_processing_size": null, "dataset_size": 7278812783.0, "size_in_bytes": 10250596365.0}}