‘Hedgehog’ Versus ‘Relatives’: Researchers Measure How the Brain Responds to Unexpected Words During Natural Speech

© iStock

Russian neurophysiologists, including researchers from HSE University, have demonstrated the feasibility of using event-related fields (ERFs) to study brain activity during natural speech perception. The researchers showed that this approach can be applied not only to individual words but also to continuous speech. Their findings indicate that words whose meanings differ significantly from the preceding context require longer processing times. The study also reveals that the brain processes function words in two stages: first, it identifies their grammatical role and then uses this information to predict the next word. The study has been published in Frontiers in Human Neuroscience.

When a person listens to someone speaking, the brain does more than simply convert sounds into words: it continuously constructs meaning, predicts subsequent words, and relates what is being heard to what has already been said. All of this happens within fractions of a second and largely without conscious awareness.

Speech perception is typically studied using event-related fields (ERFs). In such experiments, participants are presented with the same stimulus multiple times—a sound, a word, or a specially designed phrase—while the brain’s electrical potentials are recorded to capture the neural response. The signal is then averaged across all repetitions to filter out random noise and isolate the consistent response. However, such conditions differ greatly from the way people perceive live speech. Natural speech perception has been studied using complex models that reveal how the brain tracks sounds, words, and meaning within a continuous stream of speech, but these findings are difficult to compare directly with neural responses to individual words and phrases.

Researchers from HSE University and the RAS Institute of Higher Nervous Activity and Neurophysiology investigated whether the ERF approach could be adapted to the perception of an entire story. The experiment involved 27 healthy adults. While participants listened through headphones to a five-minute audio recording of a popular science text about hedgehogs, the researchers recorded participants’ brain signals using magnetoencephalography (MEG), a technique that measures the extremely weak magnetic fields generated by neuronal activity. 

The recorded signals were then averaged by semantic categories rather than individual words. The researchers compared neural responses to words that were semantically close to the preceding context with those that were more distant in meaning. The team also analysed two types of words separately: content words (nouns, verbs, and adjectives) and function words (prepositions, conjunctions, and particles). Semantic dissimilarity between words was assessed using the FastText model. 

Natalia Zhozhikashvili

'The model represents words as points in a multidimensional space and allows us to assess how close a new word is in meaning to those that have already been encountered. For example, "hedgehogs," "live," and "small" are considered semantically close to the context, whereas "white-bellied," "relatives," or "farm" are considered more distant. We compared these two groups,' explains Natalia Zhozhikashvili, Expert of the Laboratory for Cognitive Research at the HSE Faculty of Social Sciences. 

It turns out that, compared with semantically close words, those with high semantic dissimilarity elicited a late ERF effect at left temporal sensors, occurring 616–666 milliseconds after word onset. The authors suggest that this effect may be related to P600-like activity, a late component of the brain’s response that is thought to emerge when integrating a word into its context is more difficult. In studies conducted in English, the brain responded to semantically distant words earlier—around 380–600 ms after word onset. The authors assume that this difference may be related to the morphological complexity of Russian, including its use of suffixes and inflectional endings.

The brain responded to function words twice. The first peak of activity occurred between 166 and 312 ms. According to the authors, this suggests that the brain rapidly identifies the grammatical role of such words. The second peak, between 470 and 654 ms, most likely reflects predictive processing: after encountering a conjunction or preposition, the brain begins to anticipate the next word, pre-activating appropriate semantic and grammatical options. The control analysis partially supported this interpretation: content words following function words elicited an early brain response, occurring 48–98 ms after onset. However, the authors note that this effect may be linked not only to predictive processing but also to the acoustic onset of the subsequent word.

The results show that the ERF approach can be applied to continuous auditory speech in Russian. This brings language research closer to real-world conditions and may be useful for studying language disorders such as dyslexia. 

Olga Martynova

'Children with dyslexia show impaired processing of not only written but also spoken language. A non-invasive method such as MEG makes it possible to study speech processing without requiring verbal responses from participants. In the future, such approaches could be used for more accurate diagnosis of language disorders,' comments Olga Martynova, Leading Research Fellow at the HSE Centre for Cognition and Decision Making. 

The study was supported by a grant from the Russian Science Foundation.