Abstract:

Turn-taking involves the rapid alternation of speakers, with gaps between turns averaging ~200 ms. However, producing even a single word requires at least 600 ms, suggesting that speakers must prepare their responses in advance by predicting turn endings. While numerous studies have identified cues that facilitate turn-end prediction, the underlying neural mechanisms remain unclear. During turn-taking, listeners need to predict at what time the current speaker will finish their turn. In this way, turn-end prediction, in its essence, requires listeners’ temporal prediction ability. Thus, we first reviewed the studies about the neural mechanism related to time processing and prediction. Research on neural timing mechanisms distinguishes between millisecond timing, which governs event-based processing, and interval timing, which tracks longer durations (seconds to minutes). These two kinds of temporal process is related to language processing. In a proposed dual-pathway system, temporal processing is achieved in two pathways: the rapid cerebellar transmission is related to the event-based, discrete temporal processing, while the basal ganglia and striato-thalamo-cortical circuit is related to the interval-based, continuous temporal processing. The dual-pathway architecture explores how the brain processes temporal information, providing theoretical support for the neural basis of temporal prediction. However, since the model does not specify which signals enter the timing system, it remains unclear which and how speech cues are utilized to predict turn endings in conversation.

In the next part, we reviewed the cues that can be utilized by listeners to predict turn-ends. Lexico-syntactic information plays a well-established role, and despite some debate, prosodic features—particularly those at utterance-final positions—are widely recognized as predictive. Additionally, non-linguistic cues such as gaze and nodding contribute to turn-end anticipation. Although previous studies have confirmed that these cues could be utilized by speakers to predict turn-ends, they did not specify the exact role of these cues. To address this gap, we propose a temporal prediction model for turn-taking. In this model, the lexico syntactic information is transmitted linearly to the cortex via ascending pathways, which is then mapped onto meaning. The prediction of lexical-syntactic information then adjusts neural oscillations to anticipate turn endings through cortico-thalamic feedback. Meanwhile, prosodic cues are rapidly processed via the cerebellar pathway, directing cortical attention to the incoming speech signal and setting the dual-pathway system to “predictive mode”. This model integrates different types of cues for turn-end prediction and suggests a possible predictive mechanism. Within this framework, we summarize the limitations of existing research and propose further directions: future studies should 1) examine the role of different cues in dual-pathway architecture in predicting turn-ends, 2) investigate the relationship between individual temporal prediction ability and turn-taking performance, 3) use M/EEG techniques with high-temporal-resolution to study the relative weighting of lexical/syntactic and prosodic cues in turn-end prediction. 4) use free production paradigms to explore multiple cognitive processes involved in turn-taking, including comprehension, content preparation, turn-end prediction, and actual speech production.

Key words: turn-taking, temporal prediction, dual-pathway model, turn-end prediction