Postdoctoral Fellowship: SPRF: Algorithmic Models of Incremental Human Language Comprehension

Hoover, Jacob L, Montreal

Investigators

Abstract

This award was provided as part of NSF's Social, Behavioral and Economic Sciences (SBE) Postdoctoral Research Fellowships (SPRF) and SBE's Perception, Action and Cognition (PAC) programs. 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. Under the sponsorship of Dr. Roger Levy at the Massachusetts Institute of Technology, this postdoctoral fellowship award supports an early career scientist developing a framework for process-level models of human language comprehension. The ability to efficiently comprehend language, even when it is imperfect, and to adapt the amount of effort during processing is a remarkable feature of human cognition. However, most current models and algorithms for language processing don’t have this same kind of flexibility. This project aims to bridge the gap between human language behavior and computational models. By developing an understanding of algorithms that can adjust their complexity based on how well the input matches expectations we can gain crucial insights into how the human brain processes language. This research contributes to cognitive science and has potential applications in artificial intelligence, particularly in making language processing systems more efficient, adaptable, and human-like. This project will analyze incremental language processing behavior with a particular focus on imperfect input and implement computational models to describe and explain human processing behavior at an algorithmic level. This research will provide a novel framework within which to understand existing models of the relationship between incremental belief representations and processing cost, to formulate concrete testable hypotheses about computational and algorithmic models of processing. It will involve the design and collection of a corpus of text data and human reading-times, focusing on typographical errors as critical stimuli to distinguish models in this space. The project will design and implement a noisy-channel model of processing, leveraging insights from recent research in sampling-based incremental inference algorithms, which have the flexibility to demonstrate human-like time-complexity when processing textual linguistic input. The computational modeling, corpus studies, behavioral experiments, and algorithmic implementations in this project will contribute insights into the mechanisms which underlie the human capacity for accurate and efficient comprehension of linguistic input. This work has broader implications for modeling of human comprehension in cognitive science as well for artificial intelligence systems for inference and processing of linguistic data. This research will also produce open-source tools and datasets, facilitating further advancements in these fields This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.