TurnGPT: a Transformer-based Language Model for Predicting Turn-taking in Spoken Dialog

TL;DR

TurnGPT is a transformer-based model that predicts turn-shifts in spoken dialog by effectively utilizing syntactic and pragmatic context, outperforming previous baselines and capable of projecting turn completions.

Contribution

This paper introduces TurnGPT, a novel transformer-based model that improves turn-taking prediction by leveraging dialog context and pragmatic cues, with analysis of its interpretability.

Findings

TurnGPT outperforms prior baseline models in turn-shift prediction.

The model effectively uses dialog context and pragmatic cues for accurate predictions.

TurnGPT can also project turn completions, enhancing dialog understanding.

Abstract

Syntactic and pragmatic completeness is known to be important for turn-taking prediction, but so far machine learning models of turn-taking have used such linguistic information in a limited way. In this paper, we introduce TurnGPT, a transformer-based language model for predicting turn-shifts in spoken dialog. The model has been trained and evaluated on a variety of written and spoken dialog datasets. We show that the model outperforms two baselines used in prior work. We also report on an ablation study, as well as attention and gradient analyses, which show that the model is able to utilize the dialog context and pragmatic completeness for turn-taking prediction. Finally, we explore the model's potential in not only detecting, but also projecting, turn-completions.