Arabic-BERT

Pretrained BERT language models for Arabic

If you use any of these models in your work, please cite this paper:

@inproceedings{safaya-etal-2020-kuisail,
    title = "{KUISAIL} at {S}em{E}val-2020 Task 12: {BERT}-{CNN} for Offensive Speech Identification in Social Media",
    author = "Safaya, Ali  and
      Abdullatif, Moutasem  and
      Yuret, Deniz",
    booktitle = "Proceedings of the Fourteenth Workshop on Semantic Evaluation",
    month = dec,
    year = "2020",
    address = "Barcelona (online)",
    publisher = "International Committee for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.semeval-1.271",
    pages = "2054--2059",
}

Pretraining data

The models were pretrained on ~8.2 Billion words:

• Arabic version of OSCAR (unshuffled version of the corpus) - filtered from Common Crawl
• Recent dump of Arabic Wikipedia

and other Arabic resources which sum up to ~95GB of text.

Notes on training data:

• Our final version of corpus contains some non-Arabic words inlines, which we did not remove from sentences since that would affect some tasks like NER.
• Although non-Arabic characters were lowered as a preprocessing step, since Arabic characters do not have upper or lower case, there is no cased and uncased version of the model.
• The corpus and vocabulary set are not restricted to Modern Standard Arabic, they contain some dialectical Arabic too.

Pretraining details

• These models were trained using Google BERT's github repository on a single TPU v3-8 provided for free from TFRC.
• Our pretraining procedure follows training settings of bert with some changes: trained for 4M training steps with batchsize of 128, instead of 1M with batchsize of 256.

Models

	BERT-Mini	BERT-Medium	BERT-Base	BERT-Large
Hidden Layers	4	8	12	24
Attention heads	4	8	12	16
Hidden size	256	512	768	1024
Parameters	11M	42M	110M	340M

Results

Sentiment Analysis Results (F1-Score)

Dataset	Details	ML-BERT	hULMona	Arabic-BERT Base
ArSenLev	5 Classes, Levantine dialect	0.510	0.511	0.552
ASTD	4 Classes, MSA and Egyptian dialects	0.670	0.677	0.714

Note: More results on other downstream NLP tasks will be added soon. if you use these models, I would appreciate your feedback.

How to use

You can use these models by installing torch or tensorflow and Huggingface library transformers. And you can use it directly by initializing it like this:

from transformers import AutoTokenizer, AutoModel

# Mini:   asafaya/bert-mini-arabic
# Medium: asafaya/bert-medium-arabic
# Base:   asafaya/bert-base-arabic
# Large:  asafaya/bert-large-arabic

tokenizer = AutoTokenizer.from_pretrained("asafaya/bert-base-arabic")
model = AutoModel.from_pretrained("asafaya/bert-base-arabic")

Acknowledgement

Thanks to Google for providing free TPU for the training process and for Huggingface for hosting these models on their servers 😊