TurnTurk-GPT: Text-Based Turn-Taking Prediction for Turkish
What it is
This paper introduces TurnTurk-GPT, a model for predicting turn-taking in Turkish conversations by fine-tuning a Turkish GPT-2 architecture. It is evaluated on real dialogue data against rule-based and multilingual baselines, demonstrating that verb-final morphology is a critical cue for end-of-turn detection in Turkish.
The problem it solves
Existing multilingual end-of-turn detectors and zero-shot language models perform poorly on Turkish, particularly with unseen speakers or unpunctuated streaming text, due to the language's agglutinative and verb-final structure.
What you could use it for
- Detecting end-of-turn in real-time Turkish speech recognition systems
- Improving turn-taking logic in Turkish conversational AI agents
- Researching morphological cues in agglutinative languages for dialogue management
Why it matters
Enables more natural and responsive conversational AI for Turkish speakers by accurately identifying when a speaker has finished their turn, reducing interruptions and improving dialogue flow.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.