Exploring variation in sociolinguistic evaluation across human and machine talkers

Abstract Socially meaningful language variation is present in every spoken utterance. However, little research has investigated whether listeners perceive sociolinguistic variation in computer voices similarly to how they do for human voices. “Social” computers are becoming widespread in today’s social landscape, and they are designed to use many characteristics of human social communication in their speech, such as politeness and gendered language. Do listeners evaluate sociolinguistic variation as socially meaningful when used by a computer talker? The current study investigates this question by using the (ING) variable, a widespread sociolinguistic variant in American English, as a case study of sociophonetic variation. Participants listened to a sample news report read by two human and two synthetic voices that varied in their (ING) usage rate, then provided social evaluations on rating metrics of social status, appropriateness for news broadcasting, trustworthiness, and humanlikeness. We found that listeners transferred their perceptions of (ING) variation to computer voices, but also that machines’ linguistic behavior is interpreted by listeners through evolving, context-dependent social scripts. These findings have implications for both theories of social meaning and human-computer interaction.