Arabic2English - Arabic to English Translator

This is a PyTorch implementation of an Arabic to English Neural Machine Translation built using Transformers architecture (Attention Is All You Need)

Why Arabic→English is Hard

Arabic is morphologically rich — one word can express what takes a full sentence in English. It's also right-to-left with complex root-based derivation. Standard seq2seq models struggle with this — Transformers handle it better due to attention mechanisms capturing long-range dependencies.

Model Architecture

• Custom Transformer (Attention Is All You Need)
• 3 encoder + 3 decoder layers
• 8 attention heads
• Embedding size: 256
• Trained from scratch — no pretrained weights

Setup and Requirements

1. CUDA:
install CUDA before installing the required packages or check if it is already installed

2. Clone the Translate repo:

$ git clone https://github.com/Strifee/arabic2english.git

3. install requirements:

pip install -r requirements.txt

if you have problem with CUDA package try this:

conda install -q pytorch torchvision cudatoolkit=11 -c pytorch-nightly

Data

Arabic to English Translation Sentences :

Arabic to English Translation Sentences is a dataset for machine translation between English and Arabic.

Training

1. Clone the Translate repo:

$ git clone clone https://github.com/Strifee/arabic2english.git

2. Training

$ python translate.py

2. Regularization

Hyperparameters :

BATCH_SIZE = 16
learning_rate = 0.0001
num_epochs = 30

num_heads = 8
num_encoder_layers = 3
num_decoder_layers = 3

max_len= 230
dropout = 0.4
embedding_size= 256

Before regularization :

After regularization :

Results

"I'm ready" -> 'أنا مستعد'
"i'm lucky"  -> 'انا محظوظ'
"I'm sad" -> 'أنا حزين'