Your brain does not store knowledge the way a filing cabinet holds folders. Instead, meaning is assembled on the fly from a distributed network of regions, each contributing a different piece of what you know about a concept. A central hub in the temporal lobes pulls together sensory, motor, and emotional fragments scattered across the cortex, constructing a unified concept in real time. This process, which researchers call semantic cognition, involves not just the storage of knowledge but also the executive machinery that controls how and when you retrieve it, and its architecture explains everything from why certain brain diseases erase word meaning to why you struggle to name a familiar face when you are distracted.
The Hub-and-Spoke Model
The dominant framework for understanding how the brain organizes meaning is called the hub-and-spoke model. The “spokes” are sensory and motor regions distributed across the cortex. When you think about a hammer, areas involved in grasping and arm movement contribute your motor knowledge, while visual areas contribute the shape and color. When you think about a lion, visual areas light up more than motor areas, because what you know about lions is dominated by how they look rather than how you physically interact with them.
The “hub” sits in the anterior temporal lobe, particularly in a region running along the underside and front of the temporal lobes. This region does not store one type of sensory information. Instead, it merges the inputs from all the spokes into a single coherent concept. A study using magnetoencephalography (MEG) demonstrated this directly: the anterior temporal lobe responded to both the category and the level of detail required in a task, while a visual region in the fusiform gyrus responded more strongly to animals than to manufactured objects, and a motor region near the central sulcus responded more to manufactured objects than to animals. The hub cared about the concept as a whole; each spoke cared about its own sensory domain.1PLoS ONE. Oscillatory Dynamics Supporting Semantic Cognition: MEG Evidence for the Contribution of the Anterior Temporal Lobe Hub and Modality-Specific Spokes
Think of it this way: the concept “ketchup” involves knowing that it is red (visual cortex), that it is cold and smooth (somatosensory cortex), that it goes on food (functional knowledge areas), and that it comes in a specific kind of bottle (shape recognition areas). No single brain region holds the full concept of ketchup. The anterior temporal hub acts as a meeting point where all these attributes converge into something your mind treats as one idea.2PubMed Central. Controlled semantic cognition relies upon dynamic and flexible interactions between the executive ‘semantic control’ and hub-and-spoke ‘semantic representation’ systems
The Control Layer on Top of Storage
Storing knowledge is only half the problem. You also need to retrieve the right piece of knowledge at the right moment, which is harder than it sounds. The word “bat” means something very different at a baseball game than in a conversation about caves. Selecting the contextually appropriate meaning, suppressing the irrelevant one, and flexibly shifting between related concepts all require a layer of executive control that operates on top of the hub-and-spoke system.
This control system is centered on a left-hemisphere frontoparietal network, primarily the left inferior frontal gyrus and parts of the parietal cortex. When a semantic task is easy and the answer pops into your head automatically, the control network stays relatively quiet. When the task gets harder, perhaps requiring you to find a distant association or ignore a dominant meaning, the control network ramps up. Neuroimaging confirms that this semantic control network is multimodal, meaning it handles difficulty in both verbal and nonverbal knowledge retrieval, not just language.3PubMed Central. The neural correlates of semantic control revisited
The interplay between these two systems, the representational hub-and-spoke network and the executive semantic control network, is dynamic. Recent research shows that the default mode network (the set of brain regions active during rest and internal thought) and executive control regions sometimes work together and sometimes decouple during knowledge retrieval. When retrieval is effortful, both networks co-activate to support a deliberate search through memory. When access is more automatic, the default mode network can handle it with less help from the control system.4Current Opinion in Behavioral Sciences. A state-space perspective on the role of the default mode network in semantic cognition During tasks like following a story, the middle temporal gyrus, one of the key semantic regions, shows particularly strong connections with default mode areas such as the angular gyrus.5Cerebral Cortex. Functional connectivity of semantic and default mode networks during narrative comprehension
Semantic Maps Tile the Cortex
If you zoomed out and looked at the whole cortical surface, you would find that meaning is not concentrated in one hotspot. Semantic information is represented in intricate patterns covering broad regions of prefrontal cortex, lateral and ventral temporal cortex, and lateral and medial parietal cortex, and these patterns are strikingly consistent from one person to the next.6PubMed Central. Natural speech reveals the semantic maps that tile human cerebral cortex A landmark study had people listen to hours of narrative stories while in a brain scanner and then modeled how thousands of words mapped onto the cortical surface. The result was a detailed atlas showing that conceptual categories are organized into smooth gradients rather than sharp boundaries, covering much of both visual and non-visual cortex.7PubMed Central. A continuous semantic space describes the representation of thousands of object and action categories across the human brain
One finding that challenged older assumptions is that these semantic maps are widespread across both hemispheres. For decades, language-related knowledge was assumed to be a left-hemisphere affair. Brain imaging using a large vocabulary of common words has shown that the semantic system spans regions in both hemispheres, not just the left.8Nature Communications. Connecting concepts in the brain by mapping cortical representations of semantic relations
How the Brain Handles Concrete Versus Abstract Ideas
Knowing about a dog and knowing about justice feel like fundamentally different kinds of knowledge, and the brain treats them differently. A meta-analysis of neuroimaging studies found that concrete concepts, the kind you can see and touch, activate the posterior cingulate, precuneus, fusiform gyrus, and parahippocampal gyrus more strongly, regions tied to perception and mental imagery. Abstract concepts, the kind that lack a clear physical form, instead engage the inferior frontal gyrus and middle temporal gyrus more, regions associated with language processing.9PubMed Central. Neural representation of abstract and concrete concepts: a meta-analysis of neuroimaging studies
More recent work using naturalistic contexts (people reading or listening to real text, not isolated words on a screen) has refined this picture. Concrete words tend to modulate activity in sensory and motor areas, including large parts of the occipital cortex and the precentral sulcus. Abstract words, by contrast, engage regions more associated with internal and emotional processing, including medial prefrontal cortex, the angular gyri, and the superior temporal sulcus.10eLife. When abstract becomes concrete, naturalistic encoding of concepts in the brain The upshot is that your brain leans on mental imagery and sensory replay to anchor concrete concepts, while abstract concepts rely more heavily on linguistic and social-emotional scaffolding.
Category-Specific Knowledge Loss
Some of the earliest evidence that knowledge has structure in the brain came from patients who lost knowledge of one category while retaining another. After herpes encephalitis, for example, one patient lost knowledge of animals, fruits, vegetables, and flowers but had little trouble with manufactured objects. Detailed testing showed the deficit specifically affected the ability to recall the perceptual features that define living things, like shape and color, while knowledge of functional properties was preserved. When the same patient was tested on manufactured objects, problems surfaced only for perceptual details like color or sound, not for identifying the objects overall.11Cortex. Are Semantic Systems Separately Represented in the Brain? The Case of Living Category Impairment
Neuroimaging in healthy people echoes these patient findings. Processing words for living things, compared to nonliving things, produces stronger activation in the right inferior frontal gyrus, middle temporal gyrus, and fusiform gyrus.12PubMed. Activation of right fronto-temporal cortex characterizes the ‘living’ category in semantic processing The interpretation is not that the brain has a dedicated “animals drawer” and a “tools drawer.” Rather, living things depend more on shared perceptual features (shape, texture, movement patterns), while manufactured objects depend more on what you do with them. Damage to perceptual knowledge regions disproportionately affects the category that relies most on perception.
What Happens When the System Breaks Down
Two clinical conditions illustrate how different parts of the semantic system can fail independently. In semantic dementia, progressive degeneration of the anterior temporal lobes produces a gradual, relentless erosion of conceptual knowledge itself. Patients slowly lose the meaning of words and objects across all modalities: they cannot understand them when heard, read, seen, or touched.13Brain. Knowledge of famous faces and names in semantic dementia Structural brain scans highlight the temporopolar and perirhinal cortices as the focal point of damage.14The Lancet Neurology. Semantic dementia and its clinical, functional, and pathological characteristics
In semantic aphasia, by contrast, the knowledge is still there but patients struggle to access it appropriately. These patients typically have stroke damage involving the left prefrontal cortex or the temporoparietal junction, the control regions described earlier. They can often recognize a concept when given enough cues, but fail when the task demands flexible retrieval or the suppression of competing meanings.15Brain. Different impairments of semantic cognition in semantic dementia and semantic aphasia: evidence from the non-verbal domain A case-series comparison confirmed this dissociation: patients with semantic dementia and patients with stroke aphasia could fail the same tests to the same degree, but for different reasons. The dementia patients had lost the underlying representations; the aphasia patients had lost the ability to control and direct retrieval.16Brain. Semantic impairment in stroke aphasia versus semantic dementia: a case-series comparison
This distinction is not always perfectly clean. A more recent study comparing semantic dementia and comprehension-impaired stroke aphasia patients on a range of tasks found less clear-cut differences between the two groups than the theory predicts, suggesting that the boundary between damaged representations and damaged access may be blurrier than originally thought.17PubMed. Evaluating the distinction between semantic knowledge and semantic access: Evidence from semantic dementia and comprehension-impaired stroke aphasia
Testing the Hub With Magnetic Stimulation
Brain imaging shows correlations, but it cannot prove that a region is actually necessary for a task. Transcranial magnetic stimulation (TMS) can. By temporarily disrupting activity in a targeted brain area, researchers can see what happens to behavior in real time. When low-frequency TMS was applied to the anterior temporal lobes of healthy volunteers, it selectively slowed both picture naming and word comprehension without affecting non-semantic tasks of similar difficulty, mimicking the pattern seen in semantic dementia.18PubMed Central. Anterior temporal lobes mediate semantic representation: mimicking semantic dementia by using rTMS in normal participants
Further TMS work confirmed that both the left and right temporal poles contribute equally. Disrupting either one slowed semantic processing for both words and pictures, while stimulating a control site had no effect, and a perceptual task of comparable difficulty was also unaffected.19PubMed. Amodal semantic representations depend on both anterior temporal lobes: evidence from repetitive transcranial magnetic stimulation The same lab later showed that the direction of the effect depends on stimulation frequency: low-frequency TMS (which suppresses neural activity) slowed category judgments, while high-frequency TMS (which excites neural activity) actually sped them up.20PubMed Central. Enhancing vs. inhibiting semantic performance with transcranial magnetic stimulation over the anterior temporal lobe: Frequency- and task-specific effects These causal experiments are among the strongest evidence that the anterior temporal lobe hub is not merely active during semantic tasks but genuinely required for them.
The Speed of Meaning
One of the most reliable signals in brain electrophysiology is the N400, a negative voltage deflection peaking around 400 milliseconds after you encounter a stimulus. It appears whenever the brain is processing meaning, and its size reflects how unexpected or hard to integrate that meaning is. Hearing “I take my coffee with cream and sugar” produces a small N400 to “sugar,” because the word is predicted. Hearing “I take my coffee with cream and socks” produces a large N400 to “socks,” because the meaning is surprising.
The N400 is remarkably versatile. It is triggered not only by written, spoken, and signed words, but also by pictures, faces, videos, sounds, and even mathematical symbols. Its amplitude is sensitive to a wide range of factors, while its timing is strikingly constant, suggesting that the brain has a fairly fixed window for initial semantic access regardless of the format of the input.21PubMed Central. Thirty years and counting: finding meaning in the N400 component of the event-related brain potential (ERP)
Knowledge Without Sight
If so much of conceptual knowledge involves sensory features, what happens when a person has never had access to a major sense? Studies of people born blind have produced a surprising answer: the brain’s conceptual architecture develops largely intact without vision. Neural specialization for living things in temporoparietal semantic networks, including the precuneus, was observed in both blind and sighted participants. Even responses to light-related concepts (words like “rainbow” or “glow”) showed similar patterns across groups. The left middle temporal gyrus responded more to event concepts, including light events, than to entity concepts in both populations.22PubMed Central. Neural specialization for ‘visual’ concepts emerges in the absence of vision
Where the blind brain does differ is in how it handles concepts that lack any tangible sensory referent. Words for visually dominant phenomena like “rainbow” and classically abstract concepts like “freedom” both produce preferred activation in the left dorsal superior anterior temporal lobe in congenitally blind people. In this region, the less sensorially grounded a concept is, the stronger the response, suggesting that the dorsolateral anterior temporal lobe supports an amodal level of meaning that does not depend on any particular sensory channel at all.23Nature Communications. Neural representation of visual concepts in people born blind
Bilingual Brains and Shared Meaning
People who speak two languages do not maintain two separate knowledge stores, one per language. Neuroimaging of bilingual speakers performing semantic judgments in each of their languages found that the locations of brain activation were similar for both, with no differences when the two languages were compared directly. A shared frontal lobe system handles semantic analysis regardless of which language the input arrives in.24PubMed. Convergent cortical representation of semantic processing in bilinguals
That said, the sharing is not absolute. More recent and more detailed analysis has found that while semantic representations are largely shared between a bilingual person’s two languages, fine-grained differences exist in how some semantic categories are represented, presumably reflecting the cultural and experiential nuances each language encodes.25PubMed Central. Bilingual language processing relies on shared semantic representations that are modulated by each language The concept of “bread,” for instance, may carry slightly different associative weight in French versus English, and the brain’s representation reflects that, even though the core meaning is the same.
Semantic Memory and Aging
One of the more reassuring findings in this field is that semantic memory is relatively resilient to normal aging. While episodic memory (remembering specific events) declines noticeably with age, the store of general knowledge about the world tends to hold up well. A study using a rapid visual categorization task found that older and younger adults showed equivalent neural markers of semantic processing. Statistical analysis provided moderate evidence that the two age groups genuinely did not differ, rather than the study simply failing to detect a difference.26Scientific Reports. Fast Periodic Visual Stimulation indexes preserved semantic memory in healthy ageing The practical implication is that older adults who notice they are slower to retrieve a person’s name or a specific word are probably experiencing a retrieval-speed issue, not a loss of underlying knowledge.
What Large Language Models Reveal About the Brain
The rise of large language models has created an unexpected tool for studying semantic cognition. Researchers have found that the intermediate layers of these models predict human brain responses to language better than either the early layers (which encode surface-level word form) or the late layers (which encode task-specific predictions). A recent study used sparse autoencoders to decompose model representations into tens of thousands of interpretable features and found that semantic features alone accounted for about 94% of the peak ability to predict brain responses. Five semantic subcategories derived from neuroscience research mapped onto distinct cortical regions with strong alignment, suggesting that the way these models organize meaning bears a nontrivial resemblance to how the brain does it.27arXiv. Sparse Autoencoders Map Brain-LLM Alignment onto Cortical Semantic Topography
Another study used a language model to extract both concept-level and relational embeddings (representations of how two concepts relate to each other, such as a scene and an object within it). The relational information mapped onto medial and lateral parietal structures in the brain, while concept-in-isolation information mapped more onto occipital and inferior temporal regions.28Current Biology. Mapping concept and relational semantic representation in the brain using large language models These models are not replicas of the brain, but the convergences are providing researchers with new hypotheses about how meaning is structured in cortex.
Rehabilitating Lost Knowledge
When semantic cognition is damaged, whether by stroke or progressive disease, targeted therapy can make a genuine difference. In progressive aphasia, a condition where word-finding ability gradually deteriorates, intensive semantically based treatment has been shown to improve naming of trained items, with gains lasting up to a year after treatment ended. Improvement even generalized to untrained items for up to six months.29PubMed Central. Treatment for Word Retrieval in Semantic and Logopenic Variants of Primary Progressive Aphasia: Immediate and Long-Term Outcomes The treatments work by capitalizing on cognitive abilities that are still intact, reinforcing connections between concepts through repeated, structured practice rather than trying to rebuild lost neural tissue.
For stroke-induced aphasia, a randomized controlled trial of a protocol targeting phonological, lexical, and semantic deficits together found that treated patients improved in repetition, naming, word retrieval, and sentence comprehension compared to a control group, and the gains generalized to items not specifically practiced during therapy.30PubMed Central. Rehabilitation of post-stroke aphasia by a single protocol targeting phonological, lexical, and semantic deficits with speech output tasks: a randomized controlled trial The clinical picture reinforces what the basic science suggests: semantic cognition is a network phenomenon. When one node is damaged, strengthening the remaining connections can partially compensate, because meaning was never stored in one place to begin with.
How Education Shapes the Network
The structure of your semantic memory is not fixed by biology alone. How knowledge is organized appears to be influenced by how it was learned. A study comparing children educated in Montessori versus traditional schooling used word-association tasks to map the structure of each child’s semantic network. The Montessori-educated children had networks with higher clustering (meaning related concepts were more tightly interconnected) and shorter path lengths between concepts (meaning it took fewer associative steps to get from one idea to another).31Nature. Education shapes the structure of semantic memory and impacts creative thinking The researchers linked these structural differences to higher scores on creative thinking tasks, suggesting that a more interconnected conceptual network makes it easier to find remote associations, the kind of leaps that underlie creative problem-solving. Whether these structural differences persist into adulthood or are specific to the educational context studied remains an open question.