Which listening task is better at detecting phonetic imitation?
Have you ever found yourself taking on aspects of someone else’s pronunciation while speaking with them—even without intending to? This phenomenon, known as phonetic imitation, is a major focus of research in our lab.
Phonetic imitation is interesting not only as a feature of social interaction, but also because it may tell us something about how speech sounds and words are represented in the mind. Before researchers can investigate these larger questions, however, they need a reliable way to determine whether imitation has occurred.
One common method is to have listeners compare recordings made of people talking before and during an interaction with another speaker. If the recordings made during the interaction sound more like that speaker than those made before the interaction, this provides evidence of imitation. Researchers often make this comparison using a listening task called an AXB task. However, imitation effects can be extremely subtle, and listeners sometimes find the AXB task confusing.
In this study, we compared the AXB with an alternative task called the 4IAX task, which presents the recordings in pairs and may make the relevant comparison easier. Twenty-six listeners completed both tasks, using exactly the same set of recordings.
Both tasks provided evidence that imitation had occurred—but the 4IAX detected it much more clearly. Listeners selected the recording produced during imitation on 58% of AXB trials, compared with 68% of 4IAX trials. Remarkably, every one of the 26 listeners perceived more imitation in the 4IAX task, even though both tasks used the same recordings.
Performance on the two tasks was also related: listeners who showed greater sensitivity to imitation in one task tended to do so in the other. This suggests that the tasks tap into a shared perceptual ability, while the 4IAX makes it easier for listeners to use that ability.
These findings matter because the method researchers choose can affect how much imitation they detect—or whether they detect it at all. Our results suggest that the 4IAX may be the better choice, particularly when imitation is subtle. We presented this work at the conference LabPhon 20 in Montreal in June 2026, and our next step is to prepare it for publication.