Emre Sevinç
@emresevinc.bsky.social
AI | ML | Data | Architecture | Science | Math | Linux | Philosophy 👉 http://linkedin.com/in/emresevinc | https://ileriseviye.wordpress.com | https://x.com/EmreSevinc
Reminds me of "important things are seldom urgent, and what is urgent is seldom important". (On another note, being slow and smooth in order to become fast and effective heavily reminds me of martial arts)
September 19, 2025 at 12:44 PM
Reminds me of "important things are seldom urgent, and what is urgent is seldom important". (On another note, being slow and smooth in order to become fast and effective heavily reminds me of martial arts)
... then calculate its accuracy against a test data set (holdout data set).
September 19, 2025 at 8:32 AM
... then calculate its accuracy against a test data set (holdout data set).
That would be indeed a good benchmark! It should be relatively easy to compile a proper Turkish corpus with many foreign names and terms, create its 'ASCIIfied' version, then maybe use Arman's recent neural network based github.com/armish/nokta... to train a model and ...
GitHub - armish/nokta-ai: Turkish Diacritics Restoration with Neural Networks
Turkish Diacritics Restoration with Neural Networks - armish/nokta-ai
github.com
September 19, 2025 at 8:31 AM
That would be indeed a good benchmark! It should be relatively easy to compile a proper Turkish corpus with many foreign names and terms, create its 'ASCIIfied' version, then maybe use Arman's recent neural network based github.com/armish/nokta... to train a model and ...
