Turn transition in talk-in-interaction is achieved with remarkable precision, most commonly following a gap of no more than 200 ms (e.g., Stivers et al., 2009). How the precision is achieved is a complex issue given the wide range of variables co-participants to talk-in-interaction deploy to project (as speakers) and predict (as listeners) turn completion. This paper aims to contribute to a deeper understanding of one such variable used by speakers to project turn-completion: changes in word duration in turns-at-talk. As word duration varies significantly due to influences from a large number of confounds, we approach the challenges inherent in “[p]roviding robust, quantified, comparative measures of duration” (Local & Walker, 2012: 259) by fitting mixed-effects models based on naturally occurring corpus data. Contrary to previous research, which hailed the turn-final drawl as a turn-yielding cue, the models indicate that drawling, or rallentando, affects not just the turn-final syllable/word but large portions of the turn. Rallentando appears to be, not a one-off cue marking the turn’s end-point upon its occurrence, but an extended process advance-projecting the turn’s durational envelope. Also, as a graded advance-projecting resource, rallentando is in and of itself insufficient to signal turn completion reliably; listeners are likely to rely on turn rallentando in unison with other, preferably discrete cues marking the turn-completion point upon its occurrence, for “recogniz[ing] that a turn is definitely coming to an end” (Levinson & Torreira, 2015: 12) and triggering the launch of the next turn.

互动对话中的转弯转换具有很高的精度，通常不超过200 ms的间隔（例如，Stivers等，2009）。鉴于参与互动对话的参与者，部署到项目（作为演讲者）和预测（作为听众）完成转折的变量范围很广，如何实现精度是一个复杂的问题。本文旨在帮助人们更深入地理解演讲者用来预测转弯完成程度的一个变量：通话转弯中单词持续时间的变化。由于单词持续时间受大量混杂因素的影响而显着不同，因此我们通过拟合混合效应模型来应对“ [p]提供鲁棒的，持续的，量化的持续时间比较方法”（Local＆Walker，2012：259）所固有的挑战。基于自然发生的语料数据。与先前的研究相反，该模型表明转弯最后的音节是产生转弯的提示，模型表明，转弯或rallentando不仅影响转弯最后的音节/单词，而且还会影响转弯的很大一部分。Rallentando似乎不是在转弯发生时标记转弯终点的一次性提示，而是一个扩展的过程，可以预测转弯的持续时间。另外，作为分级的预先投影资源，rallentando本身不足以可靠地表明转弯已经完成；听众可能会依靠转rallentando 但需要扩展流程，才能预测转弯的持续时间。另外，作为分级的预先投影资源，rallentando本身不足以可靠地表明转弯已经完成；听众可能会依靠转rallentando 但需要扩展流程，才能预测转弯的持续时间。另外，作为分级的预先投影资源，rallentando本身不足以可靠地表明转弯已经完成；听众可能会依靠转rallentando与其他（最好是离散的）提示一致地标记转弯完成点，以“确认转弯肯定即将结束”（Levinson＆Torreira，2015：12）并触发下一个转弯的启动。转。