Summary: A high-resolution neuroimaging study shows that both the human brain and large language models (LLMs) use deeply parallel, predictive information-processing strategies to organize language. By combining continuous audiobook listening with millisecond-resolution EEG and MEG recordings, researchers demonstrated that the brain activates in advance to anticipate upcoming words: predictable words evoke quieter neural activity, while unexpected words produce stronger neural spikes.
The data indicate that, despite operating on very different physical media, biological brains and digital AI systems converge on similar internal structural strategies for representing language. These findings open new avenues for brain–computer interfaces, diagnostic tools, and personalized therapies for language-related disorders.
Key Facts
- Predictive brain confirmed: High-density EEG and MEG measurements show that the brain pre-activates language representations milliseconds before an expected word begins.
- Inverse signal intensity: Neural response amplitude scales inversely with a word’s predictability—highly probable words require less neural processing, while surprising words trigger large response spikes.
- Structural convergence: Although brains use electrochemical signaling and LLMs compute numerical probabilities on silicon, both systems appear to form highly parallel internal maps to represent language.
- Naturalistic audiobook paradigm: Instead of using isolated, artificial sentences, the study tracked continuous language processing by recording brain activity while participants listened to an audiobook.
- Clinical potential: Mapping this predictive overlap could inform diagnostics for cognitive processing deficits, enhance brain–computer interface fidelity, and guide individualized speech therapy approaches.
Source: FAU
Are humans born with innate grammatical scaffolding, or does language emerge from use and experience? This longstanding debate in linguistics has gained renewed interest with the rise of powerful AI language models—LLMs that predict the next word in a sequence. These models have sharpened questions about how language is represented and anticipated in the brain.
“In our study, we combined the continuous, natural language of an audiobook with simultaneous electroencephalography and magnetoencephalography and compared participants’ brain activity directly with the predictive probabilities generated by large language models, using a temporal resolution of mere milliseconds,” explains Dr. Patrick Krauss.
Are the brain’s predictions measurable?
The recordings show clear anticipatory activity: the brain becomes active before a spoken word actually begins. The neural response during word recognition is attenuated when the word is highly predictable, whereas unexpected words evoke much stronger neural activity. “This allowed us to demonstrate that the brain actively predicts language and that these predictions are measurable and follow patterns similar to modern language models,” says Dr. Krauss.
Language models are built from artificial neural networks—mathematical systems inspired by brain architecture. While biological nervous systems rely on electrochemical signaling, language models operate by calculating numerical values across many parameters. Despite these differences, the study found that both human brains and LLMs not only show similar prediction patterns but also appear to organize linguistic information in comparable ways.
Do our brains and AI operate on similar principles?
These results support central ideas in cognitive neuroscience and help explain why LLMs are effective in many language tasks. Similar output patterns between brains and models do not prove identical mechanisms, but they do suggest shared core information-processing principles. “The exciting question is why two such different systems converge on similar strategies for organizing language—and what the limits of that convergence might be,” notes Achim Schilling.
What comes next?
The researchers plan to test the robustness of these principles and explore practical applications. A deeper understanding of how the brain and language models represent and predict language could eventually enable new diagnostic methods, personalized therapies, higher-fidelity brain–computer interfaces, and more transparent AI systems.
Key questions answered
A: The team combined electroencephalography (EEG) and magnetoencephalography (MEG) to capture electrical and magnetic activity with millisecond temporal resolution. Participants listened to a continuous audiobook so language processing remained natural. The M/EEG sensors recorded rapid shifts in neural activity, revealing anticipatory signals that occurred before word onset and thus providing direct neurophysiological evidence of continuous predictive processing.
A: When language matches the brain’s predictions, neural responses are efficient and relatively small. An unexpected word disrupts that predictive stream, producing a larger neural response spike. This elevated activity appears to reflect an error-correction process, in which the brain updates its internal model to incorporate the surprising input.
A: No. Matching predictive behavior does not imply shared subjective experience or consciousness. The brain depends on complex electrochemical signaling and biological plasticity, while LLMs compute probability distributions and transform vectors on silicon hardware. Still, the study suggests both systems obey similar mathematical principles of information processing that lead to analogous solutions for predicting and representing language.
Editorial notes
- This article was edited by a Neuroscience News editor.
- The journal paper was reviewed in full and additional context was added by staff.
About this research
Author: Doha El Ezzi
Source: FAU
Contact: Doha El Ezzi – FAU
Image: The image is credited to Neuroscience News
Original research: Open access. “The predictive brain: Neural correlates of word expectancy align with large language model prediction probabilities” by Kölbl N, Tziridis K, Maier A, Kinfe T, Chavarriaga R, Schilling A, Krauss P. NeuroImage. DOI: 10.1016/j.neuroimage.2026.121966
Abstract
The predictive brain: Neural correlates of word expectancy align with large language model prediction probabilities
Predictive coding theory proposes that the brain continuously anticipates upcoming words to optimize language comprehension, but the neural mechanisms of this process during natural speech have been unclear. In this study, EEG and MEG data were recorded simultaneously from 29 participants while they listened to an audiobook. Predictability scores for nouns were generated using three large language models (one BERT model and two multilingual LLaMA models).
Results show that higher predictability is associated with reduced neural responses during word recognition—reflected in lower N400 amplitudes—and with increased anticipatory activity before word onset. EEG revealed increased pre-activation in left fronto-temporal regions, while MEG suggested greater sensorimotor engagement for low-predictability words, indicating a potential motor-related component to linguistic anticipation.
These findings provide new evidence that the brain dynamically integrates top-down predictions with bottom-up sensory input to facilitate language comprehension in naturalistic contexts. This work bridges computational language models and neurophysiological data, offering insights for cognitive computational neuroscience and inspiring the development of neuroscience-informed AI.