Kuwain 1.5B: An Arabic SLM via Language Injection

Summary

This paper talks about Kuwain 1.5B, a project where researchers taught an AI language model to understand and use Arabic really well, without making it forget what it already knows in other languages.

What's the problem?

The problem is that when you try to add a new language, like Arabic, to a language model that already knows other languages, it can be hard to make the model good at the new language without messing up what it knows about the old ones. Sometimes, the model gets worse at the languages it already learned.

What's the solution?

The researchers used a special technique called language injection, which lets them add Arabic to the model in a way that boosts its Arabic skills while still keeping all the knowledge it had before. This method allows the model to perform well in both the new and existing languages.

Why it matters?

This matters because it helps create smarter, more flexible AI that can support more people around the world, especially those who speak languages that aren't always included in big language models.