The ability to generate and update predictions on the fly is key to our success in comprehending language in real time. In this talk I will describe a programme of work that examines how comprehenders may incorporate semantic, syntactic, and phonological cues to update their predictions incrementally. In the first part of the talk I will present a study that examines how comprehenders may use consecutive cues to update their predictions for an upcoming noun. I will discuss convergent eye-tracking and event-related potential (ERP) evidence which suggests that comprehenders can quickly revise their predictions upon encountering disconfirming evidence, and that processing information that conflicts with an existing prediction does not incur measurable cost on subsequent semantic processing.

In the second part of the talk I will discuss on-going work in Mandarin Chinese, a tonal language, which explores the relatively understudied intersection between language prediction and phonological knowledge. In Mandarin Chinese, a syllable’s pitch contour (or lexical tone) can change depending on what follows (a phenomenon known as tone sandhi). It presents a unique opportunity to expand on existing work to examine whether and how listeners can incorporate units of information as small as a syllable’s pitch contour to anticipate upcoming words in real-time language comprehension. I will report convergent evidence from eye-tracking and mouse cursor-tracking which suggests that, even though listeners are highly sensitive to the relevant phonological patterns, they may fail to use tone information to generate predictions for upcoming words on the fly. Set against the large body of existing evidence of prediction successes, this presents a surprising case of prediction failure and may offer an opportunity for us to better understand the inner workings of the language system.