ÂÌñ»»ÆÞ

The Neural Bases of Real-Time Language Understanding

Our lab investigates the cognitive and neural mechanisms of human language. Using neurophysiological and behavioral measures together with computational techniques, we study how language is processed and learned in real time. A recurring theme of our work is how the limited processing resources of the human brain shape the way we understand and acquire language. We aim to advance basic science and to inform the treatment of disordered language and reading.

Methods: Electrophysiology of Language

Our primary tool is electroencephalography (EEG), recorded from scalp electrodes while participants read, listen to, or learn words and sentences. EEG measures the brain’s response to linguistic input with millisecond precision. That precision is essential for studying language, which typically unfolds at three to four words per second. We analyze EEG data in several complementary ways:

• averaging across many trials of the same type to obtain the event-related potential (ERP)
• applying Fourier-based techniques to measure neural oscillations (brain rhythms)
• using source-estimation methods to link scalp-recorded signals to the brain regions that generate them
• using machine learning to identify complex but systematic patterns in brain activity

Current Research

Word recognition. During normal comprehension, words arrive in rapid succession, whether through the ears or the eyes. Despite this torrent of input, comprehenders access a word’s meaning from its physical form within roughly 150–300 ms of the first sensory evidence. Word recognition relies on specialized, learned mechanisms in perceptual areas of the brain that efficiently map a word’s form onto its meaning. Critically, it also draws on higher-level knowledge of grammar, meaning, and the discourse context. This knowledge constrains how a word is interpreted, resolves ambiguities, and even allows comprehenders to predict input before it arrives. We study the rapid dynamics of word recognition by measuring well-established ERP components that index successive stages of the brain’s response to an incoming word, including the P1 (~100 ms), the N170 (~170 ms), and the N400 (~300-500 ms) components.

Sentence comprehension. Human language is inherently combinatory. Comprehenders must not only recognize individual words but also combine them into larger units whose meanings are often quite different from the sum of their parts. This requires grammatical knowledge. Grammar is what lets us understand that the man bit the dog and the man was bitten by the dog mean very different things, even though the same nouns appear in the same order. Comprehension also depends heavily on context, which is how we know that red means different things in red hair, red wine, and a red truck. All of this combinatory work must contend with pervasive ambiguity. In the boy hit the bully with the umbrella, the umbrella may be the boy’s weapon or something the bully is holding, depending on how the sentence is structured. Comprehending sentences in context is a complex dance that coordinates grammatical and semantic knowledge with memory and executive functions. We study these processes through experiments that manipulate the grammatical, semantic, and pragmatic demands placed on comprehenders. We then observe how those demands affect EEG measures with well-established functional interpretations, including the N400 (semantic access) and the P600 (structural reprocessing and error correction).

Cognitive control in language processing. When comprehension runs into trouble, such as an ambiguity, a failed prediction, or an interpretation that must be revised, it calls on executive functions. One of these is cognitive control, the general ability to detect conflict and adjust processing. Human brains have limited resources for executive functions, and those limits fundamentally shape how humans process language, in contrast to large language models with vast computational resources. To track when and how cognitive control is engaged during reading and listening, we measure frontal theta oscillations (~3–8 Hz), a well-established neural marker of cognitive control and working memory.

Language learning. Our newest line of research investigates how the brain acquires new linguistic knowledge. Whether a child is learning a first language or an adult a second, learning a language is a monumental feat of memory formation, pattern recognition, and inference. It is shaped by the nature of human brains and constrained by their limited processing resources. To study this process, we ask adult participants to learn artificial languages built from never-before-seen pseudowords arranged according to an unfamiliar grammar. Using behavioral and EEG measures, we track how brain responses change as learners acquire the grammatical and semantic aspects of a new language and how executive function mechanisms constrain how learning unfolds.