Taking turns during conversation | Research Communities by Springer Nature

When we are having a conversation with another person, we take turns with them – we switch between producing our own sentence and listening to our partner producing theirs. This process seems effortless. In fact, we often take turns with our partner without leaving long gaps between the end of their turn and the start of our own. Research suggests there is some variability in turn-taking across languages. In Japanese, the average gap between questions and answers (thought to approximate turns) is around 7 milliseconds, while in Danish it is around 400 milliseconds (Stivers et al., 2009). Importantly, this gap is typically around 200 milliseconds, regardless of the language that we speak. But although turn-taking in conversation may be fast, speaking on its own is comparatively slow. Naming a picture takes us at least 600 milliseconds (e.g., Indefrey & Levelt, 2004), while it takes us even longer to produce a sentence describing the picture (around 1500 ms; Ferreira, 1991).
How, then, are we able to quickly respond to our partner during conversation? Theories tend to agree that the answer to this question lies in pro-active response planning (e.g., Levinson & Torreira, 2015). In particular, we listen to our partner and try to guess what they are likely to say so that we can plan our own response as soon as possible. For example, if you hear a speaker start to ask Are dogs your favourite…, then you may guess that the last word is likely to be animal or pet. Based on this guess, you can begin planning your response – yes or no – before you actually hear the speaker’s final word. You will then be ready to produce your response once the speaker has finished. Under this pro-active planning account, speaking and listening overlap, thus removing the timing burden from speaking and enabling us to take turns with little gap between them.
There is much evidence to support pro-active planning. In one study (Bögels, Magyari, and Levinson, 2015), participants were faster to answer general knowledge questions when the critical information necessary for planning an answer (here, 007) was available early (e.g., Which character, also called 007, appears in the famous movies?) rather than late in the question (e.g., Which character from the famous movies is called 007?). These findings suggest that participants in the early condition used turn content (i.e., what the speaker said) to determine the appropriate answer and prepare some aspects of their response early.

What do we plan before speaking?

Although there is evidence for pro-active planning, we do not know how much of their response people plan before they speak. Planning to speak involves deciding on the message you want to convey (i.e., the content) and retrieving the words and sounds you need to convey it (i.e., the form). One possibility is that you complete both stages when you plan a response early, and so you know both the content and the form of what you want to say before you speak (we call this the early-form account). But research suggests that overlapping listening and speaking is difficult (e.g., Barthel & Sauppe, 2019) and planning both the content and the form of your response may increase this difficulty. As a result, you may reduce the overlap between speaking and listening by planning the content of your response early but its form later, right before you speak (we call this the late-form account).
Testing the accounts: Our paper
In our recently published paper, we tested between the early- and late-form accounts in three experiments using a question-answering task. We manipulated the availability of the critical content necessary for answer preparation: it was available either early or late (see Table 1). To determine whether participants planned the form of their answers early, we manipulated the length of the to-be-prepared answers, so that they were either short (single word) or long (multi-word answers).

Table 1. Example stimuli for the four conditions. The critical content necessary for planning an answer is highlighted in bold. 

Answer Length
Critical Content
Question
Expected Answer

Short
Earl
Which animal barks and is also a common household pet?  
Dog

Short
Late
Which animal is a common household pet and also barks? 
Dog

Long
Early
Which address, home to the Prime Minister, is in London?
Ten Downing Street

Long
Late
Which address in London is also home to the Prime Minister?
Ten Downing Street

We analysed how quickly participants answered the questions. Both accounts predict that participants will answer more quickly when the critical content is available early rather than late and when answers are short rather than long. The early-form account predicts an interaction between these factors. When the critical content occurs late, participants are unable to plan their answer and so they should produce a short answer quicker than a long answer.  But when the critical content occurs early, participants can plan most (or all) of the content and the form of their answer, and so the length effect should be smaller. The late-form account, in contrast, predicts that the length effect should be the same regardless of whether the critical content occurs early or late because participants will still need to plan the form of their answer in both conditions.
In all three experiments, participants answered more quickly when the critical content was available early rather than late (see Figure 1). However, we did not find convincing evidence in the individual experiments that participants were slower to produce long than short answers and there was no interaction between these two factors. Thus, the findings from the three isolated experiments did not support either the early- or the late-form account. However, a combined analysis of two of the three experiments supported the late-form account: Participants were faster to answer questions when the critical content was available early rather than late and when the answer was short rather than long, but there was no interaction between these two factors. 

Figure 1. Distribution and mean of answer times (ms) in the four conditions in all three experiments. Individual dots show individual data points. 

Together, these findings provide tentative support for a late-form account, in which participants plan the content of what they want to say early, but the form of it late. They provide insight into how we manage the cognitive demands of listening to our partner while trying to respond to them in a timely manner, thus allowing us to develop and contribute to fluent conversation.