Mutarjim: Advancing Bidirectional Arabic-English Translation with a Small Language Model
Khalil Hennara, Muhammad Hreden, Mohamed Motaism Hamed, Zeina Aldallal, Sara Chrouf, Safwan AlModhayan
2025-05-27
Summary
This paper talks about Mutarjim, a small but powerful language model designed to translate between Arabic and English in both directions.
What's the problem?
The problem is that most translation systems either need to be very large and use a lot of computer resources to work well, or they don't perform as accurately, especially with languages like Arabic that have many unique features and challenges.
What's the solution?
The researchers built Mutarjim as a compact model that still manages to beat much bigger models on standard translation tests. It also sets a new record on a tough benchmark called Tarjama-25, showing it can handle a wide range of translation tasks very effectively.
Why it matters?
This is important because it means people can get high-quality Arabic-English translations using a smaller, faster, and more efficient tool, making accurate translation more accessible for students, travelers, businesses, and anyone needing to communicate across these languages.
Abstract
Mutarjim is a compact Arabic-English translation model that outperforms larger models on established benchmarks and achieves state-of-the-art performance on a new comprehensive Tarjama-25 benchmark.