CAREER: The Effect of Trial-Level Lexical Entropy on Language Processing

Understanding and producing language is so commonplace that we may fail to appreciate the complexities involved in them. Intrinsically meaningless speech sounds and letter shapes hit our eardrums or eyeballs, and, through a series of processing steps are turned into meaningful ideas in our heads. We can do this despite the fact that the linguistic input that we receive is not perfect; we understanding language in noisy environments, when faced with idiosyncratic factors such as accents, and we can adjust to ambiguous words and structures, among many other factors. These challenges are not only due to uncertainty in the input; the human cognitive system has its own limitations, with lapses in attention or failures of memory impacting how we process language. To remain successful and efficient in tasks of language processing despite this uncertainty, the human brain generates predictions about upcoming information based on the current context and/or prior knowledge. This project investigates this prediction, specifically focusing on the process of how we pick out the intended words - based on information like meaning, part of speech, and pronunciation, among a large set of competing words. Specifically, it looks at how the overall distribution of potential upcoming words - and specifically uncertainty about upcoming words - influences our ability to perceive and produce language, measuring this distribution at the level of an individual person with their own idiosyncratic experiences of the world. In terms of broader impacts, this project builds the language science community through the development of courses engaged in language science, and training students and faculty on how to use more advanced tools to answer a variety of questions about the nature of language processing. To obtain a valid, sensitive and nuanced measure of uncertainty associated with upcoming linguistic material, the investigators measure "information entropy" for each research participant in specific sentential contexts. Information entropy reflects the uncertainty associated with the next word following a sentential context, and when measured for each participant for each context, it reflects uncertainty experienced by specific individuals in specific semantic domains, thereby providing a customized entropy measure for each individual. The investigators collect behavioral data, such as reading times, as well as brain data (ongoing brain activity commonly known as electroencephalogram or EEG), to assess how the human brain generates predictions about upcoming input. The investigators also measure other brain functions such as working memory span and attention control to examine how prediction generation interacts with those functions. The results of this project advance our knowledge about the role of prediction when humans produce and comprehend language, and how those skills interact with other critical human brain functions such as working memory and attention. This project is jointly funded by the Perception, Action and Cognition program, the Linguistics program and the Established Program to Stimulate Competitive Research (EPSCoR). 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.