Conceptual alignment in a joint picture-naming task performed with a social robot

• DOI: 10.1016/j.cognition.2022.105213
• Publication Date: 2022-07-05T05:38:09Z
• Authors: Giusy Cirillo; Elin Runnqvist; Kristof Strijkers; Noël Nguyen; Cristina Baus
• Keywords: Joint Attention; Task Analysis

Conversation entails a tight coordination between the interlocutors in terms of co-representation and linguistic alignment (e.g., word choices). In this study we investigated whether people conceptually align in a language task with a robot. 24 French native speakers alternated with an artificial partner in naming images of objects belonging to different semantic categories (e.g., mammals, clothes…). For five out of fifteen categories the robot produced the item’s category instead of the preferred basic-level name. Logistic regression models on participants' errors revealed that they adapted to the robot’s conceptual choices, and produced more category names over the course of the experiment. This pattern was most prominent for the semantic categories for which the robot had used a category name, and importantly, it applied to novel items. These results provide strong evidence for conceptual alignment affecting word choices, indicating that prototypical concepts can be overwritten in conversation to adapt to the interlocutor.