Today's large language models are trained mostly on English data. They can speak Turkish, but they do not think in it. They pass our proverbs, idioms and telling of history through the filter of another world.
Our languages are a treasure
Azerbaijani, Kazakh, Uzbek, Kyrgyz, Turkmen, Tatar and Uyghur are branches of a common root. A model that learns these languages together can translate between them instantly and keep each of them alive in the digital age.
Our proposal
Build an open dataset for Turkish and the Turkic languages through a consortium of the Turkic states' language institutes, universities and technology companies; train a shared language model on it; and use it safely in public services, education and defense.
This article in other languages: TürkçeAzərbaycanca