Researchers have created a dataset of unscripted conversations in Turkish to help improve turn-taking prediction in natural language systems. The Real-TurnTurk corpus includes synchronized video, audio, and transcriptions of dyadic interactions. A genetic algorithm is used to optimize decision rules for predicting turn transitions based on visual, acoustic, and linguistic features. --- Why it matters: This dataset matters because it addresses a specific gap in existing research on Turkish conversational dynamics, which can inform the development of more accurate natural language processing models. Source: https://arxiv.org/abs/2608.22071