BradleyVerrinder/dialectal-translation

A full-stack web app that enables bidirectional translation between Darija and English, leveraging a seq2seq LSTM encoder-decoder NLP model to accurately process and generate translations in real time. # 🗣️ Darija to English Translator Web App (Final year project at University) This is a full-stack web application built with **Django**, **Python**, **HTML**, **CSS**, and **JavaScript** that translates Moroccan Darija (a dialect of Arabic) into English. Registered users can log in to use the translator, view previous translations, and explore an intuitive, minimal interface. --- ## 📸 Preview --- ## ✨ Features - 🧠 Utilises a Seq2Seq model with a LSTM (Long Short-Term Memory) encoder-decoder architecture. - 🔐 **User Registration & Login** Create an account and log in securely using Django's built-in authentication system. - 🌍 **Darija to English Translationa and vice versa** Switch between Morrocan Arabic to English and English to Moroccan Arabic translations using a trained NLP model. - 📜 **Translation History** Logged-in users can view their previous translations. - 🎨 **Clean, Responsive Interface** Styled with HTML, CSS, and JavaScript to provide a clean and interactive user experience. --- ## 🚀 Technologies Used - **Backend:** Django (Python) - **Frontend:** HTML, CSS, JavaScript - **Authentication:** Django Auth - **Database:** SQLite (default) or MySQL (optional) - **Model:** Pre-trained NLP model (DarijaBERT or similar) --- ## 📈 Performance Whilst some sentences are translated with 100% accuracy, it can be said that the translation accuracy of my models could use some improvement. In my dissertation, I came to the conclusion that training data and computational power were the main limiting factors which contributed to the lack of accuracy in the translations. I also found that the use of a transformer model would have significantly increased my chances of accurate translations due to their ability to use self-attention to directly model relationships between any two tokens regardless of their distance in the sequence. This would be especially beneficially in my project since I found accuracy diminished rapidly in longer sentences and toward …