Linguistic alignment of redundancy usage in human-human and human-computer interaction

Abstract While speakers are theorized to ideally not include unnecessary information (redundancy) in their utterances, in reality, they often do so. One potential reason is that linguistic redundancy facilitates language communication, especially when the addressee (interlocutor) is linguistically less competent (e.g., an artificial system). In three experiments, we examined whether linguistic redundancy may arise as a result of people’s tendency to use similar linguistic features as their interlocutor does during communication (i.e., linguistic alignment) and whether redundancy alignment (if any) differs with a human interlocutor versus a computer interlocutor. We also examined whether redundancy alignment is affected by the perceived competency of the interlocutor, participants’ abilities in theory of mind (ToM), and if redundancy alignment varied across time during the experiment. Participants carried out a picture matching and naming task with a human or computer interlocutor who either always or never included redundancies in their utterances. Redundancy alignment was found across all experiments, in that speakers produced more redundancies with a redundant interlocutor compared to a non-redundant one. This alignment was also modulated by the perceived competency of the interlocutor, the time course of the interaction, and ToM abilities, suggesting that redundancy usage is affected by both automatic and strategic mechanisms of linguistic alignment.