Pretrained language models are now ubiquitous in Natural Language Processing. Despite their success, most available models have either been trained on English data or on the concatenation of data in multiple languages. This makes practical use of such models-in all languages except English-very limited. In this paper, we investigate the feasibility of training monolingual Transformer-based language models for other languages, taking French as an example and evaluating our language models on part-of-speech tagging, dependency parsing, named entity recognition and natural language inference tasks. We show that the use of web crawled data is preferable to the use of Wikipedia data. More surprisingly, we show that a relatively small web crawled dataset (4GB) leads to results that are as good as those obtained using larger datasets (130+GB). Our best performing model CamemBERT reaches or improves the state of the art in all four downstream tasks.

We introduce the first treebank for a romanized user-generated content variety of Algerian, a North-African Arabic dialect known for its frequent usage of code-switching. Made of 1500 sentences, fully annotated in morpho-syntax and Universal Dependency syntax, with full translation at both the word and the sentence levels, this treebank is made freely available. It is supplemented with 50k unlabeled sentences collected from Common Crawl and webcrawled data using intensive data-mining techniques. Preliminary experiments demonstrate its usefulness for POS tagging and dependency parsing. We believe that what we present in this paper is useful beyond the low-resource language community. This is the first time that enough unlabeled and annotated data is provided for an emerging user-generated content dialectal language with rich morphology and code switching, making it an challenging testbed for most recent NLP approaches.

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 OSCARbased and Wikipedia-based ELMo embeddings for these languages on the part-ofspeech tagging and parsing tasks. We show that, despite the noise in the Common-Crawlbased 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 crosslingual benefit of multilingual embedding architectures.

With the success of large-scale pre-training and multilingual modeling in Natural Language Processing (NLP), recent years have seen a proliferation of large, Web-mined text datasets covering hundreds of languages. We manually audit the quality of 205 language-specific corpora released with five major public datasets (CCAligned, ParaCrawl, WikiMatrix, OSCAR, mC4). Lower-resource corpora have systematic issues: At least 15 corpora have no usable text, and a significant fraction contains less than 50% sentences of acceptable quality. In addition, many are mislabeled or use nonstandard/ambiguous language codes. We demonstrate that these issues are easy to detect even for non-proficient speakers, and supplement the human audit with automatic analyses. Finally, we recommend techniques to evaluate and improve multilingual corpora and discuss potential risks that come with low-quality data releases.