Masrawy-BiLingual-v1
Masrawy-BiLingual-v1 is a bidirectional Arabic translation model designed to translate between Modern Standard Arabic (MSA) and Egyptian Arabic. This model was trained on a total of 360,000 sentences, covering both directions (MSA → Egyptian and Egyptian → MSA).
Evaluation Dataset
The model was evaluated using held-out test sets consisting of diverse, high-quality sentences in both MSA and Egyptian Arabic. The evaluation datasets included:
- 150 general conversational sentences covering everyday topics.
- 50 business and professional sentences including emails, reports, and announcements.
- 50 science and educational sentences covering basic scientific concepts and educational content.
- 50 entertainment and media sentences including movie, TV, and music contexts.
- 20 multi-sentence examples for each category (assorted topics) to test contextual translation quality.
This resulted in thousands of sentences across various domains to comprehensively assess the model's performance in realistic scenarios.
Evaluation Results
The model was evaluated on the above test sets for both translation directions. The results are as follows:
| Task | BLEU | METEOR | CHRF |
|---|---|---|---|
| MSA → Egyptian | 0.5806 | 0.7690 | 81.02 |
| Egyptian → MSA | 0.5842 | 0.7901 | 79.83 |
These metrics indicate strong performance in both directions for Egyptian↔MSA translation.
Usage
When using the model, it is important to append a postfix to the input text to indicate the desired translation direction:
- To translate to Egyptian Arabic, append
<arz>to the input. - To translate to MSA, append
<ar>to the input.
Example in Python (using Hugging Face translation pipeline)
from transformers import pipeline
model_name = "oddadmix/Masrawy-BiLingual-v1"
translator = pipeline("translation", model=model_name, tokenizer=model_name)
# Example: Translate MSA → Egyptian
input_text = "أنا أحب القراءة في المكتبة. <arz>"
translation = translator(input_text)[0]['translation_text']
print("MSA → Egyptian:", translation)
# Example: Translate Egyptian → MSA
input_text = "أنا بحب أقرا في المكتبة. <ar>"
translation = translator(input_text)[0]['translation_text']
print("Egyptian → MSA:", translation)