Conversation Research Still Requires a Human Ear

Conversations, whether in the form of small talk or a heated debate, involve a nuanced and unique exchange of words and behaviors. Recent technological advances have made it easier for conversation researchers to manage this complexity, paving the way for a deeper understanding of how humans communicate, organizational scientist Michael Yeomans (Imperial College London) and colleagues explained in a recent article for Advances in Methods and Practices in Psychological Science (AMPPS). 

Conversation research is highly dimensional, Yeomans said in an interview with APS, requiring researchers to account both for the words and phrases being used and for each individual’s conversation strategies, goals, and knowledge of the people with whom they converse. 

“The dimensionality is not just from the words—it also matters how the words are sequenced in turns, how people listen and manage topics,” Yeomans said. “All of these factors depend on the kind of person, their relationship and reputation, and also their goals for the conversation.” 

“Given all these dimensions, it is a wonder anyone converses well at all,” he added. “But it’s the structure that people use to understand one another that also allows researchers to understand dialogue data.” 

In their AMPPS article, Yeoman and colleagues F. Katelynn Boland (Columbia University), Hanne K. Collins (Harvard University), Nicole Abi-Esber (London School of Economics), and Alison Woods Brooks (Harvard University) offer a review of ongoing challenges and current best practices in the field of conversation research. 

Researchers often use text-only conversations produced by participants during experiments or harvested from existing sources such as internet forums and legal documents, Yeoman and colleagues wrote. Conducting research with voice conversations, on the other hand, is often significantly more labor intensive because researchers must perform the time-consuming work of transcribing a conversation before they can begin analyzing the content. Although human transcription services and automated speech-recognition software can help speed up this process, the output of these services still needs to be checked for accuracy and formatting issues before it can be used. 

In either case, researchers must next annotate the transcript in order to track the conversation features they will analyze. This may include generating dictionaries of how frequently words are used in the text, noting discussion topics, or identifying pauses and filler words.  

Once the analysis of these features is complete, experimenters can share the data from the  transcripts with other researchers, Yeoman and colleagues noted. Researchers must anonymize the text to remove any information that could be used to identify the participants. 

Currently, most language research in psychology has focused on correlational findings, Yeoman said, but future analyses will need to combine large datasets with traditional experimental work to generate causal findings. The value of experimental conversation research will only increase as findings begin to affect policies related to how we share information everywhere from social media platforms to healthcare settings. 

“Many conversations are being digitized and intervened upon in the modern world, and there is great opportunity to translate big data insights into actionable strategies and platform design,” said Yeomans. “We can’t make credible claims about improving conversations without that kind of experimental test.” 

Feedback on this article? Email [email protected] or login to comment.

Reference  

Yeomans M., Boland F. K., Collins H. K., Abi-Esber N., Brooks A. W. (2023). A practical guide to conversation research: How to study what people say to each other. Advances in Methods and Practices in Psychological Science, 6(4). https://doi.org/10.1177/25152459231183919