Research

Past work from the lab has contributed key insights to our understanding of language comprehension. Using controlled and naturalistic neuroimaging tasks, we have shown that a variety of linguistic processes are supported throughout a distributed brain network that is strikingly selective for language, functionally and spatially distinct within individuals from related nearby networks that support other dimensions of thinking and perceiving (Shain et al. 2020, 2022, 2023; Shain, Kean, et al. 2024), and reliably present in the brain across many task states (Shain and Fedorenko 2025). We have shown using a combination of computational modeling, neuroimaging, and behavioral experiments that real-time language comprehension is supported by separable processes of probabilistic inference and working memory (Shain, Van Schijndel, et al. 2016; Shain 2019; Shain et al. 2020, 2022; Shain, Meister, et al. 2024; Shain 2024). We have shown in computational simulations how the learning of linguistic regularities can be shaped by joint “use it or lose it” pressures to optimize both memory and prediction of the speech stream (Elsner and Shain 2017; Shain and Elsner 2019, 2020). We have shown using AI interpretability tools that strikingly abstract regularities in natural language grammar reliably control the behavior of statistical language models (LMs) with weak innate constraints (Shain, Bryce, et al. 2016; Rozner et al. 2025; Rozner, Weissweiler, and Shain 2025). And we have contributed tooling to help scientists make sense of complex cognitive processes that unfold continuously in time (Shain and Schuler 2018, 2021; Shain 2021; Shain and Schuler 2024).

Ongoing research in the lab targets (among other things) the causes and correlates of individual variability in functional brain organization, the internal structure of language-related brain areas and their connectivity to other brain systems, the relationship between prediction and memory during incremental language comprehension, the hidden functional organization of AI models and their potential utility as “laboratory animals” in the study of human cognitive functions (including language), and the clinical application of preoperative neuroimaging to improve sensor placement for language brain prostheses in people whose speech has been impaired by brain damage.