Your Brain Predicts Words Using Grammar Chunks, Not Just Next-Word Guesswork Like AI

Standing on the corner of Michigan Avenue and Randolph Street in Chicago, watching the flow of pedestrians outside the Art Institute, it’s easy to feel like the city’s rhythm is predictable—people moving with purpose, conversations flowing in familiar patterns. Yet a new study from NYU’s neuroscience team, published in Nature Neuroscience this April, reveals something counterintuitive about how we actually process language: our brains don’t work like the predictive text on our phones or the large language models powering today’s AI chatbots. Instead of simply guessing the next word based on statistical likelihood, we constantly reorganize words into grammatical chunks—phrases and clauses—before making predictions, a nuance that changes how we understand everything from language learning to AI development.

This distinction matters deeply in a place like Chicago, where over 35% of residents speak a language other than English at home, according to recent census estimates, and where institutions like the University of Chicago’s Linguistics Department and Northwestern’s School of Communication have long studied how bilingual brains process syntax differently. The NYU research, which used magnetoencephalography (MEG) to track real-time brain activity in Mandarin speakers during Cloze-style word prediction tasks, found that neural responses varied significantly depending on whether a word fit within a grammatical constituent—like a noun phrase or verb phrase—rather than just its immediate predecessor. When researchers applied the same analysis to English data from patients, the pattern held: the brain’s prediction mechanism is hierarchically structured, not flat like an LLM’s next-word matrix.

What this means for Chicago’s growing tech and education sectors is practical. Local firms developing AI-assisted language tools—whether for customer service bots in Chinatown’s little businesses or educational software piloted in Chicago Public Schools’ dual-language classrooms—might require to reconsider how they model human language prediction. Rather than optimizing solely for next-word accuracy, as many LLMs do, systems that incorporate phrase-level constraints could better mirror how actual humans anticipate speech. This aligns with ongoing work at the Toyota Technological Institute at Chicago (TTIC), where researchers explore computational models of syntax, and at the University of Illinois Chicago’s Cognitive Science program, which investigates how language processing intersects with attention and memory in urban, multilingual environments.

The study’s authors, including NYU’s David Poeppel and collaborators from the Ernst Struengmann Institute and Zhejiang University, emphasized that although LLMs and human brains both engage in next-word prediction, the similarity ends there. LLMs treat each word’s predictive context uniformly, whereas the human brain first isolates meaningful grammatical chunks—like “the red ball” or “ran down the street”—and only then evaluates which words fit best within those structures. This constituent sensitivity explains why we’re rarely thrown off by unexpected words in familiar phrases (“I sat on a…” still primes us for “chair” or “floor,” not “quantum theory”), even when the immediate predecessor offers low constraint—a nuance lost on current AI models that rely heavily on entropy and surprisal metrics alone.

Given my background in cognitive linguistics and years spent analyzing how language shapes community interaction in urban settings, if this trend impacts you in Chicago—whether you’re a speech therapist near Loyola University, a software developer in the Fulton Market district, or a parent navigating bilingual education options in Evanston—here are three types of local professionals to consider:

• Speech-Language Pathologists with neurolinguistic expertise: Gaze for clinicians affiliated with institutions like Rush University Medical Center or the Chicago Hearing Society who incorporate syntactic awareness into aphasia or language delay treatment, not just vocabulary drills.
• AI Ethics Consultants specializing in cognitive plausibility: Seek firms or independent consultants who evaluate language models not just for accuracy or bias, but for how closely their prediction mechanisms align with human psycholinguistic principles—especially those familiar with work coming out of TTIC or UIC’s Cognitive Science Lab.
• Bilingual Education Coordinators with psycholinguistic training: Prioritize professionals working in CPS’s Office of Language and Cultural Education or at nonprofit partners like Instituto del Progreso Latino who understand how grammatical chunking affects second-language acquisition and can design curricula that respect natural parsing tendencies.

Ready to uncover trusted professionals? Browse our complete directory of top-rated science and technology,brains,words experts in the Chicago area today.