How the Brain Pre-Reads Words in Milliseconds

Natural reading is an elegant dance between the eyes and the brain. Every time our eyes rest on a word, the brain is already hard at work processing not just the word in focus, but also the next one waiting just to the right. This blog post explores new emerging research with novel insights into how the brain performs this remarkable feat. The study by Wang and colleagues (2025) offers a unique window into this process, uncovering how quickly—and how deeply—the brain extracts information from upcoming words before we even look directly at them.

What makes this topic fascinating is that parafoveal processing—our brain’s ability to “preview” words in the periphery—is central to fluent reading. If this mechanism falters, reading becomes effortful, slow, and fragmented. And from a biofeedback and neurofeedback perspective, this opens a crucial conversation: how do attentional networks, oscillatory rhythms, and sensory integration support this pre-processing dance? Biofeedback and neurofeedback are ultimately training tools for improving self-regulation of neural and physiological systems, including the rhythms and cortical networks deeply involved in reading. While neurofeedback often targets patterns like alpha or theta/beta ratios to improve attentional stability, reading itself is a dynamic interplay of sensory prediction, rapid integration, and efficient neural communication.

This study addresses a long-standing question in reading research: do readers extract only the visual shapes of parafoveal words, or do they also process meaning? And if so, how quickly—and in which brain regions—does this occur? By combining magnetoencephalography (MEG) with precise eye-tracking in a naturalistic reading task, the authors reveal a compelling temporal hierarchy. Orthographic information is processed first—within roughly 68 milliseconds of fixating the preceding word—followed shortly thereafter by semantic information at around 137 milliseconds. These processes unfold across distinct neural regions, with orthographic processing emerging in the visual word form area and semantic processing in the left inferior frontal gyrus.

Seen from a broader neuroscience lens, this hierarchy reflects the brain’s incredible efficiency. It mirrors predictive processing frameworks, in which the brain uses incoming sensory information to anticipate what comes next. Neurofeedback practitioners will recognize parallels to the rapid interplay between oscillatory networks seen in attentional training, particularly the involvement of occipitotemporal and frontal circuits. Understanding how these mechanisms function during natural reading gives us a richer picture of the systems we influence when we train brain dynamics.

In the next sections, we will look closely at the methods and results of this study and consider what they reveal about reading, prediction, efficiency, and the potential for applied biofeedback and neurofeedback interventions related to reading performance and cognitive processing.

Methods

In this study, the researchers set out to understand what the brain is doing during the tiny slice of time when our eyes rest on one word but are already preparing for the next. To do this, they used a combination of two powerful tools: precise eye-tracking and magnetoencephalography (MEG).

Eye-tracking allowed the researchers to monitor exactly where each participant was looking, down to the millisecond. This was essential because parafoveal processing happens incredibly quickly and only when the reader’s gaze is on the word immediately before the target. By capturing the first fixation on this pre-target word, the researchers could identify the precise moment the brain had access to the upcoming word in peripheral vision.

MEG, on the other hand, measures the brain’s magnetic activity with exquisite temporal resolution. Unlike MRI, which shows where activity happens, MEG excels at showing when it happens. This makes it ideal for studying processes that unfold in fractions of a second—like the brain’s preview of the next word during reading.

To make sense of MEG’s rich, fast signals, the team used a technique called Representational Similarity Analysis (RSA). The idea is simple: if two words share a certain type of information—like looking similar ("writer" and "waiter") or meaning similar things ("writer" and "author")—then the brain’s response patterns to those words should also be more similar. RSA allowed the team to compare these patterns during the brief moment when the upcoming word was still in peripheral vision.

Participants read 360 short sentences, each containing target words paired with either an orthographic neighbor (one that looks similar) or a semantic neighbor (one with a similar meaning). Importantly, each of these target words was preceded by the same pre-target word within a set of six matched sentences. This clever design ensured that any similarities in brain activity were due to processing the upcoming word—not the one being directly fixated.

The researchers then analyzed the MEG activity during the pre-target fixations, looking for the earliest moments when patterns for similar words (orthographic or semantic neighbors) became more alike. These timelines offered a window into the speed and depth of parafoveal processing.

Although this study uses advanced neuroscience tools, the overall logic is intuitive: track the eyes, measure the brain’s split-second responses, and compare how similar patterns emerge when the upcoming words are related in structure or meaning. The simplified design makes the findings accessible without needing to dive into the full technical details of MEG preprocessing, spatial models, or statistical clustering techniques.

Results

The findings of this study paint a remarkably clear picture of just how quickly—and how deeply—the brain begins preparing for the next word during reading. Even before our eyes land on that upcoming word, the brain is already extracting useful information, almost as if it is running a silent preview in the background.

One of the most striking discoveries is how early orthographic information becomes available. Within about 68 milliseconds of looking at the current word, the brain begins registering the visual structure of the next one. This includes cues like letter shapes and letter order—enough to make related words (like “writer” and “waiter”) produce more similar brain responses than unrelated ones. This early processing happens in the visual word form area, a region highly specialized for recognizing written language.

Shortly afterward, the brain also begins to extract semantic information—that is, the meaning of the upcoming word. This happens at around 137 milliseconds, a timeline that is surprisingly fast for such a complex operation. In other words, even before your eyes land on a word, your brain has already begun to understand what it might mean. This semantic previewing appears to engage the left inferior frontal gyrus, a key region involved in language and conceptual processing.

Another key finding is that people who showed stronger orthographic and semantic preview effects were also faster readers overall. This suggests that fluent reading isn’t just about recognizing the word in front of us—it’s also about efficiently preparing for the next one. Stronger previewing seems to give readers a head start, reducing the load once the eyes finally land on the target word.

Taken together, these results support the idea that fluent reading depends heavily on this behind-the-scenes preview system. Rather than processing words strictly one at a time, the brain works in a layered, overlapping fashion: first capturing the visual structure of the upcoming word, then accessing its meaning—all while we’re still fixated on the previous word.

Discussion

The findings from this study offer a compelling look at how the brain orchestrates reading with astonishing speed and efficiency. Rather than processing words in a simple, step-by-step manner, the brain seems to operate like a finely tuned predictive engine—extracting information from the current word while simultaneously preparing for the next. This layered and overlapping process sheds light on why fluent reading feels so smooth and effortless, and why disruptions to these mechanisms can make reading laborious.

At the heart of these findings is the idea that the brain accesses different levels of information in a hierarchical sequence. Orthographic features are processed almost immediately, giving the brain the basic visual scaffolding of the upcoming word. Semantic information follows close behind, enabling the brain to predict meaning even before direct visual contact. This rapid transition from structure to meaning reflects an efficient division of labour across the reading network: visual areas specialize in form, while frontal language regions support higher-level interpretation.

These insights speak directly to broader themes in cognitive neuroscience, especially the concept of predictive processing. Many models propose that the brain is constantly generating expectations about incoming information, updating these predictions as new sensory inputs arrive. Reading is a perfect example: the brain uses the faint signals from peripheral vision to anticipate what is coming next, reducing the workload once the eyes reach that word. This previewing process helps explain why disruptions to attention, visual stability, or oscillatory coordination can interfere with reading fluency.

This study also highlights the brain’s astonishing temporal precision. The speed at which orthographic and semantic information emerge aligns with what we know about fast communication between occipital, temporal, and frontal regions. Rather than serial processing, the evidence here supports a partially parallel model, where different levels of information overlap in time. The brain seems to operate in a rhythm of rapid sampling, matching well with known oscillatory dynamics involved in attention and rapid perceptual integration.

Another notable aspect of the findings is the link between parafoveal processing strength and reading speed. Faster readers were those whose brains showed stronger early similarity patterns for both orthographic and semantic neighbours. This suggests that fluent reading is not only about recognizing words efficiently in the fovea; it also depends on the ability to extract meaningful information from the periphery. Readers who leverage this previewing system more effectively gain a measurable advantage in speed and fluidity.

From a practical standpoint, these insights invite deeper reflection on how we support individuals who struggle with reading. When the previewing system is inefficient—whether due to attentional instability, reduced visual sensitivity, or slower neural integration—reading can become slower and more tiring. Strategies that limit parafoveal information, such as presenting one word at a time (as seen in some rapid serial visual presentation methods), may inadvertently remove a key mechanism that fluent readers rely on.

For neurofeedback and biofeedback practitioners, this study offers an opportunity to reflect on the neural mechanisms we aim to support. Efficient parafoveal processing likely depends on stable attentional networks, smooth coordination between occipital and frontal regions, and oscillatory patterns that enable rapid transitions from visual to semantic analysis. Protocols that target attention, visual processing, or cognitive flexibility may indirectly influence the very networks that underpin fluent reading.

More broadly, this work underscores the value of integrating neuroscience findings with applied practices. As we continue to understand the brain’s capacity for rapid, predictive processing during reading, we gain insight into how training methods—whether cognitive, behavioural, or neurophysiological—may help enhance or restore this finely tuned system.

Brendan's perspective

Reading is one of those cognitive skills that looks deceptively simple from the outside. Most of us don’t think twice about it—we glance, we understand, and we move on. But under the hood, reading is a choreography of brain rhythms, finely tuned pathways, and lightning-fast transitions between sensory, perceptual, and conceptual layers. What this study offers is a rare look at that choreography, and it resonates deeply with what we see in neurofeedback practice every day.

For me, as both a scientist and a clinician, one of the most striking aspects of this research is how clearly it illustrates the importance of timing in the brain. The fact that orthographic information appears as early as ~68 ms and semantic information as early as ~137 ms reminds us that fluent cognitive processing is not just about which circuits are active—it’s about when they activate, how efficiently they transition, and how well they coordinate. Neurofeedback practitioners often talk about “training the brain to be more efficient,” but this study shows what efficiency actually looks like in real time.

And here’s where things get even more interesting. The regions highlighted in the study—the visual word form area in the left occipitotemporal cortex and the left inferior frontal gyrus—don’t operate in isolation. Their ability to communicate quickly depends on the oscillatory environment they’re embedded in. In practice, this means that a person’s peak alpha frequency (PAF) plays a central role. PAF is often considered a marker of processing speed and a general index of how quickly the brain cycles through its inhibitory–excitatory rhythms. People with a faster PAF tend to show better performance on tasks requiring rapid integration of sensory and cognitive information.

Although this study did not measure oscillations like alpha or beta directly, it’s not a stretch to imagine that individuals who showed stronger parafoveal effects—the faster readers—may also be individuals whose intrinsic oscillatory dynamics support rapid processing. A higher PAF could, theoretically, facilitate quicker transitions between orthographic analysis in occipital areas and semantic access in frontal regions. In neurofeedback training, when we help clients gently shift their alpha peak toward a more optimal range, we often see improvements in cognitive fluidity, reading comprehension, and overall mental efficiency. Studies like this one offer a mechanistic glimpse into why that may be the case.

Another dimension worth exploring is the role of beta activity, particularly in occipital and parietal regions. In some clients—especially those with reading difficulties, attentional dysregulation, or slower processing speeds—we see elevated, non-coherent beta activity in posterior regions. This kind of uncoordinated beta can create a noisy backdrop for visual and linguistic processing, making it harder for the brain to smoothly extract orthographic patterns or transition into semantic interpretation. When beta is poorly synchronised or overly generalized across occipital and parietal regions, it may disrupt the timing relationships needed for efficient parafoveal previewing.

In practice, this could show up as difficulty scanning text, inconsistent reading rhythm, or a sense of “stuttering” through sentences—not because the child or adult can’t decode, but because their visual and attentional networks are not synchronizing efficiently. Neurofeedback practitioners working with dyslexia, ADHD, or processing speed challenges likely recognize this pattern. In many of these cases, training to reduce excessive, uncoordinated beta—while simultaneously stabilizing alpha—can yield surprising improvements in reading fluency and comfort. This study helps explain why: by calming the noisy beta landscape, the brain may regain the capacity to perform fast, clean transitions between orthographic and semantic layers during reading.

What’s particularly valuable here is the reminder that the brain’s reading network is not a rigid pipeline. It’s flexible, adaptive, and highly dependent on individual state variables—attention, arousal, and oscillatory tone. This is where neurofeedback shines. While controlled laboratory studies must constrain variables tightly to isolate mechanisms, clinical practice is the opposite: we individualize. We adapt. We look not at one brain region or one frequency band, but at the entire dynamic pattern a client brings into the room.

For example, consider a child struggling with reading fluency. In traditional academic approaches, we focus on phonics, decoding, or comprehension strategies. But from a neurofeedback standpoint, we might also ask: What does their posterior alpha look like? Is there excess beta over occipital sites? Is their PAF unusually slow? Are attentional networks engaging efficiently when shifting from one fixation to the next? These questions connect directly to the kinds of micro-timings highlighted in this study.

Similarly, adults with attentional fatigue or cognitive overload often describe difficulty “pacing” their reading. They can read the words, but they cannot keep the rhythm. For these individuals, training to support alpha stability and reduce noisy beta often restores a sense of flow—not because we’ve taught them to decode, but because we’ve optimized the neural conditions that support parafoveal preview and rapid integration.

One of the recurring themes in my own clinical experience is that improvements in reading often emerge indirectly through training aimed at broader regulatory systems. When we support clients in building a more stable alpha peak, reducing posterior beta noise, or improving frontal–occipital coherence, gains in reading fluency, attention, and processing speed often follow. The brain doesn’t compartmentalize these systems the way our research designs do. It operates as an integrated network, and parafoveal processing is a perfect example of this integration.

This study also highlights an important nuance: real reading is different from controlled reading tasks. Natural reading is fast, dynamic, and context-rich. Neurofeedback training must reflect that complexity. Rather than assuming that improving reading requires targeting one narrow region or frequency, we can view reading fluency as emerging from a broader network-level efficiency. That perspective aligns perfectly with the individualized, whole-brain approach that experienced clinicians take.

Finally, findings like these should encourage us to remain humble and curious. We are working with systems that operate at millisecond precision and involve delicate flows of information across multiple brain regions. Our job is not to impose rigidity but to create conditions that allow the brain’s natural efficiency to re-emerge. When clients become more synchronized—when their oscillatory rhythms settle into a state of flexible stability—skills like reading often improve as a natural consequence.

In short, this research beautifully complements what we see clinically. It helps explain why optimizing oscillatory dynamics, particularly alpha and beta activity in posterior regions, can support reading performance. And it reminds us that the brain’s reading network is not just a cognitive apparatus—it’s a living, time-sensitive, rhythm-based system. When we train the rhythms, the reading often follows.

Conclusion

The study reviewed here reveals something profound about the way we read: fluent reading is not a simple matter of sequentially processing one word at a time, but a dynamic, multilayered act of prediction. The brain quietly previews the next word even before the eyes land on it, extracting first its visual form and then its meaning in a matter of milliseconds. This fast, hierarchical mechanism helps explain why reading can feel so fluid when everything is working well—and so effortful when it is not.

The findings also highlight just how coordinated the reading network must be. Regions at the back of the brain rapidly decode letter patterns, while frontal language areas begin preparing for meaning almost immediately afterward. These processes overlap, interact, and unfold with incredible temporal precision. The fact that individuals with stronger parafoveal processing were also faster readers underscores the role of this previewing system in supporting reading fluency.

For practitioners in neurofeedback and biofeedback, these results reinforce something we see repeatedly in clinical practice: that cognitive performance—reading included—depends on the brain’s ability to regulate timing, rhythm, and coordination across multiple regions. A stable oscillatory environment, efficient transitions between networks, and the ability to sustain attention all contribute to this delicate previewing mechanism. When these systems are dysregulated, reading can become slower, more effortful, or more fragmented. When they are optimized, reading often becomes smoother and more efficient.

This study gives us a richer understanding of why neurofeedback interventions that target attentional stability, posterior alpha tuning, or reductions in uncoordinated beta activity often yield improvements in reading fluency, even when reading itself is not the primary focus of training. By supporting the foundational rhythms and communication pathways involved in rapid sensory–cognitive transitions, we indirectly strengthen the neural systems that make efficient reading possible.

Ultimately, the work of Wang and colleagues reminds us that efficient reading is more than a cognitive skill—it is a neurophysiological achievement. It reflects the brain’s capacity to anticipate, integrate, and coordinate information at extraordinary speeds. These findings inspire both clinicians and researchers to continue exploring how training the brain’s regulatory systems can support this remarkable capacity.

As we continue learning from studies like this one, one message becomes clear: when we nurture the brain’s rhythm and efficiency, we nurture the foundations of comprehension, learning, and connection. And in doing so, we help unlock the natural fluency that makes reading one of the most extraordinary human abilities.

References

Wang, L., Frisson, S., Pan, Y., & Jensen, O. (2025). Fast hierarchical processing of orthographic and semantic parafoveal information during natural reading. Nature Communications, 16, 8893. https://doi.org/10.1038/s41467-025-63916-y