Bilingual Brains Read Language Labels One Word at a Time, Study Suggests

For the millions of people who move effortlessly between two languages every day, the act of reading can feel deceptively seamless. A bilingual commuter scanning a multilingual sign, or a student flipping between English and Dutch textbooks, appears to process each word instantly, switching linguistic gears without any visible hesitation. But beneath that apparent fluency lies a puzzle that has occupied cognitive psychologists for decades: when a bilingual reader looks at a line of text containing words from two different languages, does the brain determine which language each word belongs to one word at a time, or can it classify several words simultaneously?

A new study by Aybora Koc, Agnesa Xheladini, and Mathieu Declerck of the Vrije Universiteit Brussel, published in the journal Attention, Perception, & Psychophysics, offers a carefully measured answer. The researchers asked Dutch–English bilinguals to perform a lexical decision task embedded in a flankers paradigm, a classic experimental setup in which a central target word is flanked by other words that participants are asked to ignore. By manipulating how long the words stayed on screen, the team could control whether participants had enough time to process the words sequentially or were forced to rely on parallel extraction. The results point to a striking conclusion: language membership information, the mental tag that tells the brain which language a word belongs to, appears to be extracted in parallel across multiple words only in a limited way, and only when there is sufficient time to do so.

The logic of the experiment hinges on a phenomenon called the flanker-language congruency effect. In previous work, researchers including Declerck and colleagues had shown that when the flanking words around a target are drawn from the same language as the target, bilingual participants respond faster and more accurately than when the flankers belong to the other language. That difference in response times, a congruency effect, is taken as evidence that the brain registers the language identity of the flankers even though they are irrelevant to the task. If no such effect appeared, it would suggest that the flanking words were never processed deeply enough for their language membership to be extracted at all.

What makes the new study distinctive is the way it separates parallel from serial processing using viewing time. In one condition, the stimulus display was presented for only 50 milliseconds, a duration so brief that participants could not realistically scan the target and then the flankers one after another. Any congruency effect observed under this constraint would have to reflect genuinely parallel extraction of language information, because the visual input vanished before a sequential sweep through the words could be completed. In a second condition, the display lasted 170 milliseconds, long enough to allow at least a rudimentary sequential reading of the central word and its neighbors. This timing manipulation transforms the flanker task into a kind of temporal microscope for mental architecture.

The outcome was clean and interpretable. With 170 milliseconds of viewing time, the researchers found a reliable congruency effect: participants were slower to make lexical decisions about the central target when the flankers were in a different language than when they were in the same language. With only 50 milliseconds, that effect disappeared entirely. The pattern suggests that when time is abundant enough for sequential processing, the language tags of multiple words end up being registered, producing cross-language interference. When time is compressed to the point where only truly parallel extraction could operate, the language membership of the surrounding words apparently fails to be picked up, at least in a way that influences behavior.

The authors interpret this as evidence for limited parallel extraction of language membership information across multiple words. The phrasing is deliberately cautious. It does not mean the bilingual brain is incapable of any parallel word processing; a rich body of research, including work on parafoveal preview benefits during natural reading and studies of syntactic extraction within 50 milliseconds, indicates that readers glean considerable information from several words at once. What the new findings suggest is that one specific type of information, the language label attached to each word, is not automatically and fully harvested in parallel from every word in view. Instead, its extraction appears to depend on processing time in a way that is consistent with more serial or attentionally restricted mechanisms.

This nuance matters because theoretical models of bilingual word recognition disagree about how language membership is represented and used. Prominent frameworks, such as the bilingual interactive-activation model and its developmental extensions, propose that words from both languages sit in a single integrated lexicon and that language nodes help regulate competition between them. Under such views, language membership information is intimately woven into the earliest stages of visual word recognition, and one might expect it to be extracted from any word that receives enough visual input. The new results complicate that expectation: even when flankers are clearly visible and geometrically adjacent to the target, their language identity does not automatically modulate performance unless processing time permits a deeper pass.

The study also speaks to a broader and sometimes heated debate in the science of reading: whether visual word recognition is fundamentally serial or parallel. Some researchers, drawing on eye-tracking, dual-task, and neuroimaging evidence, have argued that lexical access itself operates through a serial bottleneck, so that only one word at a time reaches full recognition even though lower-level visual and orthographic information may be gathered from several locations simultaneously. Others, working with parallel graded attention models such as OB1-reader and with flanker paradigms showing parallel semantic and syntactic effects, contend that multiple words can be processed to fairly deep levels at the same moment. The present findings carve out a middle position for one particular kind of information. Orthographic and even syntactic cues may survive very brief exposures, but the language tag, at least in this paradigm, does not.

Methodologically, the experiment was rigorous in the ways modern cognitive psychology demands. The researchers used mixed-effects regression models with appropriate random effects for both participants and items, ensuring that the congruency effects were not driven by a few unusual words or a subset of participants. The data, stimuli, and analysis code have been made openly available through the Open Science Framework, allowing other researchers to scrutinize and reanalyze the results. The study was funded by the Strategic Research Program of the Vrije Universiteit Brussel and approved by the university’s ethics committee. The authors note that the study was not preregistered, a transparency disclosure that has become standard practice in the field.

For everyday bilinguals, the practical implications are subtle but intriguing. The findings suggest that the brain does not maintain a constant, effortless surveillance of which language every visible word belongs to. Instead, language tagging may engage when attention and time allow, which could help explain why bilingual readers occasionally experience momentary hesitations at language boundaries in mixed-language text, such as code-switched sentences, even when both languages are highly familiar. As cities, classrooms, and media feeds become increasingly multilingual, understanding the temporal mechanics of how the reading brain sorts words by language is more than an academic curiosity. It is a window into how one of humanity’s most common cognitive skills actually works, one millisecond at a time.

Subject of Research: Parallel versus serial extraction of language membership information during bilingual visual word recognition

Article Title: Can bilingual readers process language membership of multiple words in parallel?

Article References: Koc, A., Xheladini, A., & Declerck, M. (2026). Can bilingual readers process language membership of multiple words in parallel?. Attention, Perception, & Psychophysics, 88(7), Article 207. https://doi.org/10.3758/s13414-026-03334-4

Image Credits: AI Generated

DOI: 10.3758/s13414-026-03334-4

Keywords: bilingualism, visual word recognition, flanker task, language membership, parallel processing, serial processing, lexical decision, language switching, psycholinguistics, eye movements, reading, cognitive psychology