Parsing continuous speech: the role of neuronal oscillations and sentential context
Abstract
The characterization of the representations and computations underlying spoken languageprocessing remains a formidable challenge for cognitive neuroscience. The experimental research program outlined in this proposal aims to address some of core issues that arise in this context from a new angle: recent neurophysiological, psychophysical, and computational research suggests that cortical neuronal oscillations may provide the infrastructure to process critical temporal attributes of continuous speech. Moreover, the benefit for recognition afforded by the contextual information contained in continuous everyday speech likely intersects with these hypothesized brain mechanisms. The PI will test whether a model of speech comprehension based on cortical oscillations that is neurobiologically realistic (Giraud & Poeppel 2012), computationally explicit (Ghitza 2011), and theoretically well motivated (Poeppel et al. 2008) can account for some of the central questions that continue to confront research in this domain. To this end the PI will develop new corpora to evaluate the role of context and new materials to better assess the role of temporal information in spoken language understanding. The knowledge acquired by this research program will provide a framework to address a major weakness common to current, state-of-the- art human language technologies: compared to humans, their accuracy degrades sharply when the input differs from the type of data on which the technology was developed or trained, e.g., when language, dialect, genre and domain are changing, or when environmental conditions areworsening.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 09, 2018
- Source ID
- FA95501810055
Entities
People
- Oded Ghitza
Organizations
- Air Force Office of Scientific Research
- Boston University
- United States Air Force