In an opinion article in Trends in Cognitive Sciences, and Stephen Levinson from the and argue that we need models of language that account for how we process language in face-to-face, multimodal dialogue.