Chiebuka and mentors in JML

A predictive coding model for online sentence processing.

Congratulations to recent undergraduate alum Chiebuka Ohams, along with his mentors Sathvik, Philip, and former postdoc Shohini Bhattasali (Toronto), whose paper, "A predictive coding model for online sentence processing," just appeared in the Journal of Memory and Language. The paper, abstracted below, argues "that sequential predictive coding models are a valuable complement to surprisal theory as a route to progress on process-oriented theories of language comprehension."

At Maryland Chiebuka was a double major in Linguistics and Computer Science. He is now at MIT as an RA in Ev Federenko's Lab and a Technical Associate in the McGovern Institute for Brain Research.

Computational approaches to prediction in online sentence processing tend to be dominated by computation-level surprisal theory, offering few insights into underlying cognitive mechanisms. Conversely, predictive coding is an algorithmic theory grounded in neuroscience, but it has rarely been employed in the study of language processing, in part because its areas of application have not involved sequential processing. Building on a recently proposed temporal predictive coding model, we present what is to our knowledge the first exploration of sequential predictive coding in broad-coverage online sentence processing. We investigate our model at non-toy scale using naturally occurring language, establishing its cognitive validity via comparison with reading times, and we link measurable aspects of the model to cognitive discussions of mechanism for prediction in language processing. Our results suggest that sequential predictive coding models are a valuable complement to surprisal theory as a route to progress on process-oriented theories of language comprehension.