Speakers adapt speech based on percieved miscommunication

Abstract

Recent work suggests that speakers can exaggerate their speech based on perceived communicative success (Buz et al., 2016). However, the flexibility and sophistication of this speech adaptation is relatively under-explored. We thus investigate whether this adaptation can be responsive to context-specific demands (e.g., exaggerating speech in contexts with increased potential for miscommunication), whether adaptation can target specific phonetic/acoustic properties (e.g., specifically exaggerating aspects of speech that may reduce miscommunication), and whether this adaptation involves inference about the potential causes of miscommunication. We investigate these questions in a web-based spoken communication game where we manipulate speakers' perceived communicative success. We find that speakers “smartly” adapt to miscommunication so as to increase speakers' likelihood of subsequent communicative success. The results argue for a speech production system that incorporates past experience and can adapt fine-grained properties of speech to better achieve communicative goals.