This model translates English text into colloquial Egyptian Arabic. It is fine-tuned from the Helsinki-NLP/opus-mt-tc-big-en-ar base model on a specialized dataset of English-Egyptian Arabic translation pairs, ensuring natural, idiomatic, and culturally appropriate translations reflective of everyday Egyptian speech.

Model Details

Model Description

This is a specialized neural machine translation model that translates from English to Egyptian Arabic (Masri/Egyptian Colloquial Arabic). Unlike standard Arabic translation models that produce Modern Standard Arabic (MSA), this model is trained to generate authentic colloquial Egyptian Arabic with natural speech patterns, slang, and tone appropriate to contemporary Egyptian communication.

Model type: Neural Machine Translation (Seq2Seq Transformer)
Language(s) (NLP): English (en), Egyptian Arabic (ar)
License: Apache 2.0
Finetuned from model: Helsinki-NLP/opus-mt-tc-big-en-ar

Uses

Direct Use

This model is designed for translating English content into colloquial Egyptian Arabic. Suitable applications include:

Social media content localization
Dialogue systems and chatbots
Educational materials adaptation
Content creation for Egyptian audiences
Travel and hospitality applications
Business communication in Egyptian contexts

Out-of-Scope Use

Formal or official document translation (prefer professional human translation or MSA models)
Legal or medical documents requiring precision and formal language
Highly technical or specialized terminology translation without domain-specific fine-tuning

How to Get Started with the Model

Use the code below to get started with the model:

python

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("Omar-youssef/english-egyptian-arabic-translator")
model = AutoModelForSeq2SeqLM.from_pretrained("Omar-youssef/english-egyptian-arabic-translator")

def translate(text):
    inputs = tokenizer.encode(text, return_tensors="pt")
    outputs = model.generate(inputs, num_beams=4, early_stopping=True)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

print(translate("We haven't seen each other in a long time. We should catch up soon."))
# Output: إحنا مش بنشوف بعض من فترة طويلة. المفروض نلحق ببعضنا قريب.

print(translate("It's getting cold outside, so take a jacket with you."))
# Output: الجو طلع برد بره، فاخد جاكيت معاك.

Training Details

Training Data

The model is trained on the Omar-youssef/English-Egyptian-Arabic-Translation dataset, specifically designed for English-to-Egyptian Arabic translation. The training data includes:

Training Procedure

Training Hyperparameters

Training regime: (bf16)
Base Model: Helsinki-NLP/opus-mt-tc-big-en-ar
Learning Rate: [5e-6]
Batch Size: [16]
Epochs: [5]

Evaluation

Testing Data, Factors & Metrics

Metrics

BLEU Score: [0.4111]

Technical Specifications

Model Architecture

Architecture: MarianMT (Encoder-Decoder Transformer)
Base: Helsinki-NLP/opus-mt-tc-big-en-ar
Parameters: ~500M
Task: Sequence-to-Sequence Translation

Compute Infrastructure

Hardware: Kaggle GPU (T4 )
Framework: Hugging Face Transformers
Training time: ~60 minutes

Citation

If you use this model, please cite the base model and dataset:

bibtex

@misc{omar-youssef-egyptian-translator,
  author    = {Omar Youssef},
  title     = {English to Egyptian Arabic Translation Model},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/Omar-youssef/english-egyptian-arabic-translator}
}

Model Card Contact

For questions or feedback, open a discussion on this model's Hugging Face page.