Minnesota Natural Language Processing Seminar Series: Investigating Language in the Brain Using Artificial Neural Networks
The Minnesota Natural Language Processing (NLP) Seminar is a venue for faculty, postdocs, students, and anyone else interested in theoretical, computational, and human-centric aspects of natural language processing to exchange ideas and foster collaboration. The talks are every other Friday from 12 p.m. - 1 p.m. during the Fall 2021 semester.
This week's speaker, Greta Tuckute (MIT), will be giving a talk titled "Investigating language in the brain using artificial neural networks."
The human language system allows us to infer meaning from text and speech. The unique human ability to comprehend language depends on a left-lateralized fronto-temporal brain network that responds robustly and selectively to linguistic input. A big open question in cognitive science and neuroscience concerns the organization of the language system. Until recently, we had no computationally precise models that could serve as quantitative hypotheses for how core aspects of language might be implemented in the mind and brain. However, artificial neural networks (ANNs) for language have suddenly achieved impressive performance on a wide range of language tasks – prompting the question of whether ANNs can serve as the first computationally precise models of how the human brain may solve the same tasks.
In my talk, I will discuss: i) ANNs as models of sensory systems, ii) Methodological approaches and assumptions underlying the use of ANNs as models of language processing, iii) Findings from a large-scale investigation1 of 43 diverse ANN language models as models of human neural (fMRI/ECoG) and behavioral responses. In brief, we found that match-to-brain correlated with next-word prediction performance of ANNs (but not performance on other GLUE benchmarks) and we thus claim that a drive to predict future inputs may shape human language processing.
Schrimpf, M., Blank, I.A., Tuckute, G., Kauf, C., Hosseini, E.A., Kanwisher, N.G., Tenenbaum, J.B., & Fedorenko, E. (2021). The neural architecture of language: Integrative modeling converges on predictive processing. Proceedings of the National Academy of Sciences, 118.
Greta Tuckute is a PhD student in Brain and Cognitive sciences at MIT working with Dr. Ev Fedorenko. She obtained her BSc and MSc degrees from University of Copenhagen in Denmark. She is now working in the intersection of neuroscience and AI, and is interested in exploiting artificial neural networks to understand how language is processed in the mind and brain.