TurnTurk-GPT predicts Turkish turn-taking from unpunctuated text
Original title: TurnTurk-GPT: Text-Based Turn-Taking Prediction for Turkish
The fine-tuned Turkish GPT-2 beats a multilingual detector, but ties a zero-shot model on unseen speakers.
TurnTurk-GPT is a text-based Turkish turn-taking predictor made by fine-tuning a Turkish GPT-2 with the TurnGPT objective on synthetic and real dialogue. Tests use word-level points on unpunctuated, lower-cased text that approximates streaming speech recognition, with recording-disjoint and speaker-disjoint folds. It beats the multilingual detector in both settings, but matches the zero-shot Turkish model on unseen speakers. Suffix removal points to verb-final morphology as the main cue, and the model and protocol are released.
- Removing turn-final suffixes cuts average precision by up to 44%.
- Synthetic dialogue alone produced the weakest trained model.
- Zero-shot and multilingual baselines stay near chance during speaker pauses.
- Evaluation uses nested thresholds and paired cluster bootstrap tests.
From the original article
1 min read at huggingface.co
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.
By Ahmet Tuğrul Bayrak, Fatma Nur Korkmaz, Nur Yağmur Tuncer · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.