Neural Mechanisms of Probabilistic Prediction in Language Comprehension

Morgan Emily I, Somerville MA

Investigators

Abstract

This award was provided as part of NSF's Social, Behavioral and Economic Sciences Postdoctoral Research Fellowships (SPRF) program. The goal of the SPRF program is to prepare promising, early career doctoral-level scientists for scientific careers in academia, industry or private sector, and government. SPRF awards involve two years of training under the sponsorship of established scientists and encourage Postdoctoral Fellows to perform independent research. NSF seeks to promote the participation of scientists from all segments of the scientific community, including those from underrepresented groups, in its research programs and activities; the postdoctoral period is considered to be an important level of professional development in attaining this goal. Each Postdoctoral Fellow must address important scientific questions that advance their respective disciplinary fields. As the world becomes more interconnected, it is increasingly important to be able to communicate with people from different linguistic backgrounds. Learning to communicate within new linguistic norms is not merely a feature of early childhood language learning but continues throughout the lifetime. It is thus essential to study how speakers incorporate statistical knowledge in new linguistic environments. This project takes a first step towards understanding the neural mechanisms that enable comprehenders to adapt to new linguistic norms. In particular, this project studies how comprehenders use their statistical knowledge about the language and the world (i.e. what things people tend to talk about, and what language they use to do so) to make predictions about upcoming words, and how these predictions adapt to a changing environment. The project focuses on three neural mechanisms: 1. Retrieval of words from long-term memory. 2. Selection between competing alternatives by activating one word and suppressing others. 3. Conflict resolution when a prediction is violated by the actual linguistic input. The project will further our understanding of language comprehension by developing computational models of how these neural mechanisms adapt to changing statistics in the linguistic environment (e.g. an interlocutor who tends to say very predictable versus very unpredictable things) and testing these models against electrophysiological data. Ultimately, the project will elucidate how these neural mechanisms underlie language comprehension and how they allow for lifelong language learning. In order to comprehend language rapidly in a noisy and ever-changing world, comprehenders must leverage their statistical knowledge about the language to resolve uncertainty. This project studies how neural mechanisms make probabilistic predictions about upcoming words and how these predictions adapt to changes in the linguistic environment. The project combines computational modeling with cognitive neuroscience to make quantitative predictions about the role of distinct mechanisms in language processing, with a focus on testing the hypothesis that prediction in language comprehension is a rampant, probabilistic process. First, the project will develop a hierarchical generative model of probabilistic prediction in language processing to make quantitative predictions about distinct neural mechanisms using principled information theoretic measures. The project will then investigate how the brain can implement this computational theory using relatively recently discovered event-related potential (ERP) components, beyond the time window of the classic N400 effect, crucially using these ERP components to further our functional understanding of how these neural mechanisms implement algorithmic processes hypothesized by the computational framework. Finally, the project will use the computational framework and further ERP experiments to investigate whether and how these mechanisms adapt to environments in which unexpected events occur frequently. These experiments will elucidate the neural mechanisms that underlie predictive processes in language comprehension, and how such mechanisms allow for lifelong language learning. The project will advance our understanding of language comprehension by unifying disparate theories from computational psycholinguistics and cognitive neuroscience. The project includes substantial methodological innovation beyond the current state-of-the-art, using generative models to make quantitative predictions both about neural mechanisms broadly and about learning curves during language adaptation. The investigators will also develop methods for trial-by-trial ERP data analysis which are particularly powerful for studying ongoing adaptation in changing environments.