TurnTurk-GPT: Text-Based Turn-Taking Prediction for Turkish

A paper on arXiv (2610.42010v1) introduces TurnTurk-GPT, a Turkish text-based turn-taking predictor built by fine-tuning a Turkish GPT-2 model with the TurnGPT objective on synthetic and real dialogue. Evaluated on unpunctuated, lower-cased text approximating streaming speech recognizer output, TurnTurk-GPT outperformed a widely used multilingual end-of-turn detector under both recording-disjoint and speaker-disjoint folds, but beat a zero-shot Turkish language model only when test speakers appeared in training; removing suffixes of the turn-final word cut average precision by up to 44%, versus 2% for a sentence-internal word, locating the Turkish end-of-turn cue in verb-final morphology. The model and evaluation protocol are publicly released.

arXiv:2610.42010v1 Announce Type: new Abstract: This paper presents TurnTurk-GPT, a text-based turn-taking predictor for Turkish obtained by fine-tuning a Turkish GPT-2 model with the TurnGPT objective on synthetic and real dialogue, and evaluates it on real Turkish conversation against rule baselines, a zero-shot Turkish language model and a widely used multilingual end-of-turn detector. Evaluation uses word-level decision points on unpunctuated, lower-cased text approximating the output of streaming speech recognisers, recording-disjoint and speaker-disjoint folds, nested threshold selection and paired cluster bootstrap tests. TurnTurk-GPT outperforms the multilingual detector under both recording-disjoint and speaker-disjoint evaluation. It outperforms the zero-shot language model only when the test speakers occur in training; on unseen speakers the two are statistically indistinguishable. Removing the suffixes of the turn-final word reduces AP by up to 44%, against 2% for a sentence-internal word, locating the Turkish end-of-turn cue in verb-final morphology. Synthetic dialogue alone yields the weakest trained model, and when a speaker pauses the zero-shot and multilingual baselines remain close to chance. The model and the evaluation protocol are publicly released.