Turkish researchers create the first dataset for teaching AI to summarize educational videos

Researchers built a dataset of 82 Turkish course videos with 3,281 human-written summaries, then created an automated method to extract the most-agreed-upon content across all those summaries. This produces a single gold-standard summary that machines can learn from, making it possible to train AI systems to summarize videos in Turkish instead of just English.

Educational video summarization has been a solved problem in English for years, but only because researchers had datasets to train on. Turkish didn't have one. This dataset closes that gap for a language with roughly 90 million speakers. The real signal is methodological: the consensus approach they developed (AutoMUP) is cheap enough to apply to other Turkic languages without starting from scratch.