Semantics: How Language and the Brain Create Meaning

Semantics is the study of meaning in language, and it reaches into nearly every corner of how humans think, communicate, and make sense of the world. It covers the meaning of individual words, how words combine into larger meanings, and how context reshapes what a sentence actually conveys. But semantics is not just an abstract branch of linguistics. It shows up in brain scans when you read an unexpected word, in the way toddlers latch onto new vocabulary after hearing it once, in courtroom arguments over what a statute “really means,” and in the failures of AI systems that can string together grammatically perfect sentences without fully grasping what they are saying.

How Meaning Lives in Your Head

One of the first things semantics research established is that word meanings are not stored like dictionary entries with crisp boundaries. Instead, your mental categories have fuzzy edges organized around central examples. A robin feels like a more typical “bird” than a penguin does. A desk chair feels like a more typical “chair” than a beanbag. This is the prototype effect, and it is not a cultural quirk. Research with members of a Stone Age culture that had no existing color or geometric-form vocabulary found the same pattern: categories built around perceptually natural prototypes were easier to learn than categories where those prototypes were placed at the edges, and people consistently picked prototypical examples as the best representatives of a category even when the categories were brand new to them.

1Cognitive Psychology. Natural categories

Words also carry multiple related meanings, a phenomenon called polysemy. “Run” can mean jogging, operating a machine, or a sequence of events. How your brain handles those related senses depends on the type of word. Research on nouns, verbs, and adjectives found that for nouns and verbs, people mentally store the literal and metonymic senses of a word together, while the metaphorical sense sits apart. Adjectives work differently: their metonymic senses overlap with both the literal and the metaphorical, creating a more blended mental picture. So the word class itself shapes how tightly related meanings cluster in your mind.

2PubMed Central. The Mental Representation of Polysemy across Word Classes

Meaning Is Grounded in the Body

A powerful idea in modern semantics is that word meanings are not abstract symbols floating free of experience. Instead, understanding a word activates some of the same brain regions you would use if you were actually perceiving or doing what the word describes. This is the core claim of embodied semantics, and a growing body of neuroscience supports it. A consensus paper gathering evidence across many studies concluded that language representation is intrinsically tied to our physical interactions with the world and that this connection leaves measurable traces in the brain during both language processing and language learning.

3PubMed Central. Brain Signatures of Embodied Semantics and Language: A Consensus Paper

A striking demonstration of this comes from research on how new word meanings take root. When adults learned brand-new spoken words for visual objects, hearing those words afterward triggered activity in the primary visual cortex, the earliest stage of visual processing. The activation was category-specific: the particular pattern depended on what the word referred to. This was the first direct evidence that learning a new word forges a link between the sound of the word and sensory information in the parts of the brain that handle raw perception, not just in some higher-level language center.

4PubMed Central. Semantic Grounding of Novel Spoken Words in the Primary Visual Cortex

This grounding matters because it means meaning is not simply a lookup table in one dedicated brain region. It is distributed across the sensory and motor systems you used when you first encountered the thing a word refers to. The smell of coffee, the motion of throwing, the texture of sandpaper: each of these leaves its mark on what the associated word means to your brain.

The Brain’s Meaning Machinery

Even though meaning is distributed, the brain does have regions that act as coordination hubs. The anterior temporal lobes, located at the front tips of the temporal cortex, appear to function as central organizers for semantic knowledge. Damage to these areas produces devastating loss of word meaning, while neuroimaging studies in healthy people consistently show them lighting up during tasks that require understanding concepts regardless of whether the input is a word, a picture, or a sound. Research describes the anterior temporal lobes in both hemispheres as a supramodal hub for verbal and nonverbal semantic information.

5Journal of Neuroscience. The Anterior Temporal Semantic Hub in the Left and Right Hemispheres: A Unitary System or Two Separate Systems for Meaning?

A study of patients with brain damage mapped where reduced connectivity in the brain predicted problems with semantic tasks. The left anterior temporal lobe and the bilateral anterior cingulate cortex emerged as hubs whose disruption predicted deficits in both general and modality-specific semantic performance.

6PubMed Central. Left Anterior Temporal Lobe and Bilateral Anterior Cingulate Cortex Are Semantic Hub Regions: Evidence from Behavior-Nodal Degree Mapping in Brain-Damaged Patients

There is also a well-known electrical signature of meaning processing that researchers have been studying for decades. When you read a word that does not fit the context of a sentence, your brain produces a characteristic negative voltage shift roughly 400 milliseconds later, known as the N400. The more unexpected the word, the bigger the signal. This response is not limited to written words: it appears for spoken and signed words, for pictures of faces and objects, for sounds, and even for mathematical symbols. Its amplitude changes with how well something fits semantically, while its timing stays remarkably constant.

7PubMed Central. Thirty years and counting: finding meaning in the N400 component of the event-related brain potential (ERP)8Electroencephalography and Clinical Neurophysiology/Evoked Potentials Section. Topography of the N400: brain electrical activity reflecting semantic expectancy

The N400 has become one of the most widely used tools for studying how the brain builds meaning in real time. It tells researchers, for instance, that your brain is constantly making predictions about what comes next in a sentence, and that the work of integrating meaning happens almost immediately, not after you have finished reading the whole thing.

How Children Crack the Code

Children face an extraordinary semantic challenge: they must figure out what thousands of words mean, often from ambiguous situations where a single utterance could refer to any number of things in the environment. One of the strategies young children use is called fast mapping, the ability to form a rough link between a new word and its referent from just one or two exposures. Research showed that two-year-olds can infer the referent of a novel noun from a single ambiguous exposure and retain that mapping over time, despite the task being designed to be genuinely difficult.

9PubMed. Rapid fast-mapping abilities in 2-year-olds

Another strategy involves mutual exclusivity: if a child already knows a word for one object, and they hear a new word while looking at that object and an unfamiliar one, they tend to assume the new word refers to the unfamiliar object. This helps narrow down possibilities fast. But the reliance on mutual exclusivity is not universal. A study comparing monolingual and bilingual preschoolers found that both groups performed similarly during the initial fast-mapping stage, but monolingual children relied on mutual exclusivity significantly more than bilingual children did.

10PubMed. Language-Fair Fast Mapping and Mutual Exclusivity Tasks for Mono- and Bilingual Preschoolers

This makes sense when you think about it: if you grow up hearing two languages, you know perfectly well that one object can have two names. Mutual exclusivity becomes a less reliable assumption. The finding highlights how the semantic strategies children use are shaped by their linguistic environment, not just by brain maturation.

Can Language Change What You See?

One of the most debated questions in semantics is whether the language you speak actually shapes how you perceive the world, not just how you talk about it. The strongest evidence comes from color. Russian has separate basic terms for light blue and dark blue, treating them as distinct categories the way English treats green and blue. English lumps both under “blue.”

A neural network model trained to simulate this difference found measurable perceptual consequences. In the Russian-trained model, two shades of green that shared the same label were perceived as highly similar, while two shades of blue that received different labels (one “light blue,” one “dark blue”) were perceived as more different from each other. In the English-trained model, both blue shades, sharing one label, were treated as more similar to each other. The language labels themselves shifted internal representations of color similarity.

11PubMed Central. How language modulates color perception in a brain-constrained deep neural network

Behavioral research with human participants largely agrees. A study measuring how language affects color cognition found evidence that a core universal component of color perception exists, but it is modulated by a small relativist influence from language. The differences showed up even in a visual search task that tapped into pre-attentive processes, suggesting the language effect kicks in early, not just when people are consciously deliberating about colors.

12British Journal of Psychology. Linguistic relativism and colour cognition

So the answer to the old Sapir-Whorf question, “Does language shape thought?”, is not a simple yes or no. Perception has a strong universal core, but language nudges it at the margins, particularly at category boundaries where one language draws a line and another does not.

What Happens When Meaning Breaks Down

The clinical side of semantics reveals what meaning processing looks like when it fails. The most dramatic example is semantic dementia, a form of progressive brain disease that gradually erodes a person’s knowledge of what words and concepts mean while leaving other cognitive abilities largely intact, at least initially. A case report described a patient who presented with complaints of memory loss but whose core problem turned out to be a sweeping semantic loss across all types of nouns, common and proper, concrete and abstract. Despite this, the patient continued to live independently and maintain daily activities, underscoring that semantic knowledge can collapse while other mental faculties remain functional.

13PubMed Central. Distinguishing Semantic Variant Primary Progressive Aphasia from Alzheimer’s Disease

Alzheimer’s disease also affects semantic memory, but in a qualitatively different way. Research directly comparing the two conditions found that patients with semantic dementia showed no priming effect for any type of semantic relationship, meaning their underlying semantic network was substantially degraded. Alzheimer’s patients, by contrast, still showed priming for broader categorical relationships, indicating a partially intact network rather than wholesale destruction.

14PubMed Central. The underlying mechanisms of semantic memory loss in Alzheimer’s disease and semantic dementia

Personal experience also plays a protective role in semantic dementia that it does not play in Alzheimer’s. Patients with semantic dementia were more likely to recognize and remember famous names that had high autobiographical significance to them personally, compared to equally familiar names that had low personal significance. Patients with Alzheimer’s disease and patients with amnesia from other causes did not show this preference. The finding suggests that in semantic dementia, personal experience can temporarily shore up meaning that the disease would otherwise erase, while in Alzheimer’s, that autobiographical lifeline is not available.

15PubMed. The contribution of autobiographical significance to semantic memory: evidence from Alzheimer’s disease, semantic dementia, and amnesia

The Gap Between What You Say and What You Mean

Semantics has a sibling discipline, pragmatics, that deals with the meaning that arises from context rather than from the words themselves. The philosopher Paul Grice drew a foundational distinction between what is said, which is closely tied to the conventional meanings of the words, and what is conversationally implicated, which listeners infer from the fact that an utterance was made in a particular context.

16Philosophy Compass. Conversational Implicatures (and How to Spot Them)

If a friend asks whether a restaurant is good and you respond, “Well, it’s cheap,” you have not said the food is bad. Semantically, your sentence is just about price. But pragmatically, the listener draws a clear inference: if the food were good, you would have said so. That inference, the implicature, carries real communicative weight despite being technically unsaid. This is how everyday conversation works much of the time. People routinely mean more, less, or something different from what their words literally encode, and listeners unpack those layers almost effortlessly.

The semantics-pragmatics boundary is not just an academic curiosity. It shows up whenever there is a dispute about what someone “really meant.” Did a politician’s promise technically cover a situation the audience assumed it did? Does a product label’s wording actually claim what consumers believe it claims? Meaning never lives in the words alone; it is always a collaboration between what the language encodes and what the context supplies.

Sound, Feeling, and the Texture of Words

Beyond reference and logic, words carry an expressive texture that semantics has only recently begun to formalize. Synesthetic metaphors, phrases like “a warm color” or “a sharp taste,” pair senses in ways that feel natural despite being literally impossible. Research found that the meanings of synesthetic metaphors generally align with crossmodal correspondences, the systematic links people draw between different senses (high pitch feels “bright,” low pitch feels “dark”). However, when a synesthetic metaphor becomes conventional and widely used, its meaning can drift away from those correspondences. The alignment was stronger on emotionally loaded dimensions like good-bad and clean-dirty than on more neutral ones like low-pitched versus high-pitched, suggesting that emotional associations mediate the connection between sensory metaphors and crossmodal perception.

17PubMed Central. The Meanings of Synesthetic Metaphors Closely Align With Crossmodal Correspondences

Actual synesthetes, people who involuntarily experience one sense when stimulated by another, appear to be more sensitive to these crossmodal links, but only for sound-symbolic associations rather than low-level sensory ones. This supports the view that the effect is more about high-level cognitive processing than about the raw senses bleeding into each other.

18PubMed Central. Synesthesia strengthens sound-symbolic cross-modal correspondences

Machines and the Meaning Problem

Computational semantics has been a cornerstone of natural language processing for decades. One of its workhorse ideas is distributional semantics: representing a word’s meaning by the company it keeps. If two words tend to appear in similar contexts across a large body of text, they are treated as having similar meanings. This approach has been used for everything from search engines to medical records. In clinical text analysis, for instance, distributional representations of words, measured by how similar their usage patterns are, have been used alongside traditional features like dictionary matching and pattern recognition to extract medical concepts from health records.

19PubMed Central. Enhancing clinical concept extraction with distributional semantics

Large language models like GPT-4 and its successors have pushed distributional methods to a remarkable scale, producing text that reads fluently and often sounds knowledgeable. But their handle on meaning has limits. When tested on composite tasks that combine two or more simpler operations, these models showed a split. They performed reasonably well on simpler composite tasks that apply distinct operations to different parts of the input, and scaling up the model improved performance. But on more complex compositions that required chaining multiple reasoning steps, performance dropped, and making the model larger did not reliably help.

20arXiv. Do Large Language Models Have Compositional Ability? An Investigation into Limitations and Scalability

Compositionality, the ability to combine known meanings into new ones, is one of the hallmarks of human semantic competence. The fact that it remains a stumbling block for even the largest AI models suggests that statistical patterns of word co-occurrence, however powerful, do not fully capture what humans do when they understand language.

Semantics in the Courtroom

Legal disputes frequently hinge on what a word means in ordinary usage, and corpus linguistics has recently entered the courtroom as a tool for settling those arguments empirically rather than by judicial intuition. A study analyzing the term “child” across three corpora representing different registers of American English found a strong preference for interpreting the word as referring to born individuals. Linguistic evidence supported extending the term to fetuses only in marked, non-default contexts. The study highlighted the tension that arises when a court’s interpretation of a word diverges from how that word is actually used in everyday speech.

21PubMed Central. Statutory Interpretation and Corpus Evidence: Tensions Between Linguistic Reality and Legal Meaning

This kind of work illustrates how semantics moves from abstract theory to practical consequence. When a statute uses a word and two parties disagree about what it covers, the question is genuinely a semantic one, and answering it with data from millions of real-world usages is a very different approach than relying on a judge’s personal sense of what the word “obviously” means. Legal corpus linguistics is still young and controversial, but it represents one of the most direct applications of semantic research to high-stakes decision-making.

Do Animals Have Semantics?

Whether meaning is uniquely human is a question that pushes semantics to its edges. Many animal species produce calls that correlate with specific situations, vervet monkey alarm calls being the classic example. But correlation is not the same as reference. A genuinely referring signal would need to be produced intentionally and directed at an audience, not simply triggered by a stimulus the way a smoke alarm is triggered by heat.

Recent philosophical work has argued that at least some animal signals clear this bar. Drawing on field studies of chimpanzees, one analysis makes the case that chimpanzees’ “alert hoo” vocalizations can plausibly be interpreted as intentional, human-like reference. These calls are produced flexibly, adjusted for audience, and directed at specific external objects or events in a way that goes beyond a simple stimulus-response pattern.

22Synthese. A case for animal reference: beyond functional reference and meaning attribution

If this interpretation holds up, it suggests that some of the raw materials for semantics existed before human language evolved, and that the leap to full-blown human meaning was not a single all-or-nothing step but a series of incremental advances. The debate is far from settled, but it reframes the question from “do animals have language?” (they clearly do not, in the full human sense) to “do animals have the beginnings of meaning?” That is a harder question to answer and, for understanding how human semantics came to exist, arguably a more interesting one.