Semantic Memory In Psychology

Semantic memory is a type of long-term declarative memory that refers to facts, concepts and ideas which we have accumulated over the course of our lives (Squire, 1992).

Semantic memory generally encompasses matters widely construed as common knowledge, which are neither exclusively nor immediately drawn from personal experience (McRae & Jones, 2013).

Examples of Semantic Memory

• Recalling that Washington, D.C., is the U.S. capital and Washington is a state.
• Recalling that April 1564 is the date on which Shakespeare was born.
• Recalling the type of food people in ancient Egypt used to eat.
• Knowing that elephants and giraffes are both mammals.

These examples share one property. You can know them without recalling when or where you learned them. This is called “noetic” awareness. It means knowing without re-living the original episode, which is what separates semantic memory from episodic memory, tied to a specific time and place (Tulving, 1972).

Semantic memory also stores conceptual relationships, such as knowing a canary is a bird and a bird is an animal. It stores word meanings too, along with abstract or formal knowledge such as the rules of arithmetic (McRae & Jones, 2013).

History and Background

Endel Tulving first proposed the term “semantic memory” in 1972. He drew on earlier work by Reiff and Scheerer (1959), who had distinguished between different forms of long-term memory. Building on this, Tulving separated episodic memory from what he called semantic memory. The new term stuck.

The distinction proved influential.

Tulving described semantic memory in a chapter published within the edited volume Organization of Memory (Tulving & Donaldson, 1972). He described it as “words and verbal symbols, their meanings and referents, the relations between them, and rules, formulas, or algorithms for manipulating them” (Tulving, 1972, p. 386).

Tulving (1984) later refined the distinction. He compared semantic and episodic memory on how each operates, the type of information each processes, and how each applies to real-world knowledge versus personal experience.

What Research Suggests

Since Tulving’s proposal, many experiments have tested his hypothesis.

Behavioral Dissociation Evidence

Jacoby and Dallas (1981) tested 247 undergraduate students in a study with two phases: a perceptual identification task and an episodic recognition task.

Jacoby and Dallas used the experimental dissociation method, and their results showed a clear difference in performance between the semantic and episodic tasks, supporting Tulving’s hypothesis.

This early behavioral test predates the neuroimaging evidence below by more than two decades.

Neuroimaging Evidence

The development of functional magnetic resonance imaging and positron emission tomography has opened up new ways to explore semantic memory’s neural organization (Eiling, Chrysikou & Thompson-Schill, 2013). These neuroimaging methods can reveal brain activity during tasks ranging from matching pictures to naming objects.

The findings are striking.

This work suggests semantic memory involves several anatomically and functionally different systems. No single brain region plays a privileged role in retrieving or representing semantic knowledge.

Each attribute-specific system is instead linked to a sensorimotor modality, along with related properties within that modality.

Location matters.

Neuroimaging studies suggest semantic memory can be categorized into types of visual information, such as motion, form, size and color.

For instance, Thompson-Schill (2003) has proposed that the left lateral temporal cortex retrieves knowledge of motion, the parietal cortex retrieves knowledge of size, and the bilateral or left ventral temporal cortex retrieves knowledge of form and color.

Networks spanning the premotor, parietal, and ventral and lateral temporal cortex appear to form semantic representations organized by category and attribute. This does not rule out nonperceptual conceptual knowledge in more anterior regions of the temporal cortex.

Lexical retrieval may be tied to posterior language regions, while semantic processing within the temporoparietal network may instead be linked to the anterior temporal lobe.

Episodic vs Semantic Memory

Semantic memory is focused on facts, ideas, and concepts. Episodic memory, on the other hand, refers to the recalling of particular and subjective life experiences.

While semantic memory embodies information generally removed from personal experience or emotion, episodic memory is characterized by biographical experiences specific to an individual.

Hence, the latter involves actual events which had transpired at specific moments in one’s life.

Semantic Memory	Episodic Memory
General knowledge	Personal experiences
Facts, concepts, meanings	Specific events and episodes
Objective and impersonal	Subjective and personal
Not tied to a specific time or place	Temporal and contextual details
Shared and common across individuals	Unique to an individual’s experiences
Used for understanding language and concepts	Used for remembering autobiographical events
Less susceptible to degradation or forgetting	More susceptible to degradation or forgetting
Examples: knowing the capital of a country, understanding mathematical concepts	Examples: recalling a specific birthday celebration, remembering a vacation trip

Semantic memory refers to general knowledge and facts, while episodic memory involves personal experiences and specific events tied to a particular time and place.

Semantic Network

A semantic network is a cognitively based graphic representation of knowledge that demonstrates the relationships between various concepts within a network (Sowa, 1987). A taxonomic hierarchy may order the organization of a semantic network’s arcs and nodes.

A node is a symbol that represents a specific word, feature, or concept, whereas an arc is a symbol that stands for a two-place relationship between nodes (Arbib, 2002). Unlike neural networks, semantic networks are unlikely to use distributed representations for concepts.

The structure varies.

A semantic network can be either a directed or an undirected graph (Sowa, 1987). The vertices represent concepts. The edges stand for the semantic relations between them, connecting and mapping different semantic fields.

Models of logical comprehension, discourse and artificial intelligence often portray knowledge this way (Barr & Feigenbaum, 1982). Natural language processing applications such as word-sense disambiguation and semantic parsing also employ semantic networks (Hoifung & Domingos, 2009; Sussna, 1993).

Three models compete here. Each explains how this network is organized and searched.

The Hierarchical Network Model

Aim. Collins and Quillian (1969) proposed a hierarchy of concepts. The hierarchy obeys cognitive economy: each property is stored once, at its most general level. They predicted that verification time should rise with the number of levels between a concept and its stored property.

Method. Participants judged simple sentences. Examples included “A canary can sing” and “A canary is an animal.” Their reaction times were recorded throughout. The statements were built so the property or category sat zero, one, or two levels above the subject noun.

The results were clear.

Results. Verification time rose with each extra level. “A canary is a bird” was confirmed faster than “A canary is an animal”, and each additional level added roughly a constant delay.

Conclusion. Extra levels added cost, exactly as predicted. This supported a hierarchy in which properties are stored with maximum economy at the highest applicable node. The model later ran into typicality effects it could not explain, which is why the spreading-activation and feature-comparison models below replaced it.

The Spreading-Activation Model

A revision followed.

Collins and Loftus (1975) revised this account. They replaced the strict hierarchy with a network organized by semantic relatedness, where the length of the arc between two concepts reflects how closely related they are.

When a concept is accessed, activation spreads outward along its arcs. It travels faster along short, related arcs.

A node that is already partly activated needs less extra input to be recognized. Closely related words are therefore recognized faster than distant ones.

This mechanism explains semantic priming: recognizing a word is faster when it follows a related word (see below). It also explains typicality effects, since typical category members sit on shorter arcs to their category and so are pre-activated more strongly than atypical members.

This flexibility is a strength and a weakness, as the Critical Evaluation section below explains.

The Feature-Comparison Model

Smith, Shoben and Rips (1974) rejected stored links altogether. On their account, a concept is a set of semantic features, split into defining features (necessary for category membership) and characteristic features (typically but not necessarily true).

Verifying a statement runs a two-stage comparison. First, an overall similarity check gives a fast answer when the two concepts are very alike or very different. Only a middling case is slow. It triggers a second stage that compares defining features alone.

Typicality decided the outcome.

A robin shares many characteristic features with “bird” and is verified quickly. A penguin shares fewer, so it forces the slower second stage. This reproduces the typicality effect that defeated the strict hierarchical model.

The model explains typicality elegantly. But it struggles to say which features truly count as defining for most everyday concepts, a problem philosophers had already raised about natural categories.

Semantic Priming

Semantic priming is a related effect. People recognize or respond to a word faster when it follows a related word. Reading NURSE, for example, speeds up recognition of DOCTOR compared to an unrelated prime such as BREAD.

This is the clearest such evidence. It shows semantic memory is organized by relatedness, the same mechanism the spreading-activation model above was built to explain.

Meyer and Schvaneveldt (1971): The Founding Demonstration

Aim. Meyer and Schvaneveldt (1971) asked whether retrieving one word from memory makes it easier to retrieve a related word straight after.

Method. Participants completed a lexical-decision task. On each trial, they saw a pair of letter strings and pressed “yes” only if both were real English words. Some “yes” pairs were related (BREAD-BUTTER); others were unrelated (NURSE-BUTTER).

Results. Participants were roughly 85 milliseconds faster to respond “yes” when the two words were related than when they were not.

Conclusion. Accessing one word automatically speeds access to related words. This semantic priming effect became one of cognitive psychology’s most widely used measures. The standard modern version simply presents a single prime shortly before a single target word.

This straightforward, well-controlled paradigm gave researchers a reliable way to probe how strongly any two concepts are connected.

Automatic Versus Strategic Priming

Priming has two separate sources. Neely (1977) varied the delay between a prime and a target, and also manipulated what participants expected to see next.

One effect is fast and automatic. It is driven purely by relatedness and present even at short delays. A second effect is slower and strategic, driven by conscious expectation.

It builds up only at longer delays. It can even slow down an unexpected target.

At long delays, an unexpected target could even be recognized more slowly than an unrelated word, a sign that expectation itself, not just relatedness, was now shaping retrieval.

In short, two systems are at work. The automatic effect is the direct behavioral signature of spreading activation. The strategic effect reflects a separate, controlled retrieval process, a distinction later research would develop into a full account of semantic control (see Contemporary Research below).

Semantic Satiation

Semantic satiation is the temporary loss of a word’s meaning after it is repeated or studied for an extended time. The word or phrase briefly seems meaningless to the person repeating it (Das, 2014).

Leon James Jakobovits first introduced the term in 1962. Across several experiments, he showed how different cognitive exercises could produce this effect.

Jakobovits (1962) proposed that verbal repetition activates a neural pattern in the cerebral cortex corresponding to the word’s meaning. Rapid, repeated firing of this pattern causes reactive inhibition, which weakens the pattern’s intensity with each repetition.

Examples of semantic satiation include the following:

• Repeating words verbally and then grouping them into ideas
• Rating figures which are shown repeatedly within a short time
• Repeating a set of numbers aloud and adding them soon afterward

Semantic satiation has been used to treat phobias, through cognitive activities that induce it as part of systematic desensitization (Jakobovits, 1966). Repetition-induced satiation has also been used to reduce speech anxiety, by dulling the negative emotion tied to a feared word.

Later work went further.

Tian and Huber (2010) later challenged Jakobovits’s neural-fatigue account. They argued satiation instead reflects weakening of the link between a word’s form and its meaning, not fatigue of the concept itself. This associative account has informed later research into word learning, reading, and multilingualism (Fishman, 2014).

Brain Regions Related to Semantic Memory

Hippocampal and Distributed Views

The neuroscience behind semantic memory has long been debated. Many clinicians and researchers have held that the brain systems storing semantic memory are the same ones that store episodic memory (Vargha-Khadem, 1997).

The debate has two sides.

On this view, the hippocampal formation and the medial temporal lobes play a vital role in storing semantic memory. The hippocampal formation encodes memories, and the cortex stores them once encoded.

The hippocampal formation includes more than the hippocampus itself. It also includes the entorhinal and perirhinal cortices, together known as the para-hippocampal cortices. Some researchers suggest that encoding may therefore be physiologically based outside the hippocampus proper.

Support for this idea comes from a study of amnesiacs who kept intact semantic memory despite hippocampal damage. Their para-hippocampal cortices had been spared.

Other researchers disagree. Some hold that semantic memory resides in the temporal neocortex. Others hold that it is distributed across all brain regions instead (Vargha-Khadem, 1997; Binder & Desai, 2011).

On the distributed view, the memory of a dog might stem from both visual and auditory cortex. A dog’s bark may be registered in the latter, while its visual features enter the former.

The Hub-and-Spoke Model

A synthesis emerged.

Recent research points to a combined account. It is called the “hub-and-spoke” model, and it combines both views above. Conceptual knowledge is neither wholly localized nor wholly distributed, but both at once.

Much of what a concept “means” is stored near the sensory and motor systems used to experience it, its “spokes”. Knowing what a dog looks like draws on visual cortex. Knowing how to use a hammer draws instead on motor areas (Thompson-Schill, 2003; Binder & Desai, 2011).

Concepts blend both.

But features alone cannot explain everything. A banana and a canary are alike because both are yellow; a banana and a baseball bat are alike because both are long and hand-held. Generalizing across features like this needs a hub that integrates every modality.

The anterior temporal lobes are proposed as this hub, binding the spokes into concepts that generalize beyond any one sense (Patterson, Nestor, & Rogers, 2007).

Semantic Dementia

Hodges, Patterson, Oxbury, and Funnell (1992) described semantic dementia. It is a progressive condition marked by shrinking vocabulary and a severe loss of single-word comprehension.

The deficit is selective. Grammar stays intact. Everyday and autobiographical memory are relatively spared, and non-verbal perceptual skills remain strong, yet knowledge of what things are becomes radically impoverished.

The cause is a progressive atrophy of the anterior temporal lobes, the hub proposed above. The hub, not the spokes, is damaged.

Because damage at the hub produces a broad loss of concepts while other memory systems stay intact, semantic dementia is a key piece of evidence for the hub-and-spoke model.

Because it selectively targets meaning while sparing memory for events, semantic dementia also helps clinicians tell it apart from Alzheimer’s disease, where episodic memory usually fails first.

Category-Specific Deficits

Semantic breakdown can also be limited to one category of knowledge. Warrington and Shallice (1984) studied patients recovering from herpes-simplex encephalitis who were far more impaired at identifying living things and foods than inanimate objects.

The reason was structural.

They explained this with category-by-modality organization. Knowledge of living things leans on sensory features, such as what a leopard looks like. Knowledge of tools leans instead on functional features, such as what a hammer is for. The two rarely overlap.

Damage that strips out one feature type impairs whichever categories rely on it most. The pattern held across both visual and verbal tests, which ruled out a simple explanation based on one sense alone.

Later work catalogued several such patterns, arguing that only an account built from many features could explain them all.

Critical Evaluation

The idea of semantic memory as a distinct system is well supported. Several questions remain, though, about how it works and where exactly it lives in the brain.

Strengths

Semantic memory has several things going for it as a scientific construct.

• Convergent support: The construct is backed by reaction-time studies, priming experiments, brain-damage case studies, and neuroimaging, a rare degree of agreement for a cognitive construct.
• Explanatory reach: Network and activation models account for a wide range of findings, including priming, typicality effects, and the speed of everyday fact retrieval, with a single mechanism.
• A clear neural signature: Conditions such as semantic dementia give semantic memory an identifiable, selectively vulnerable basis in the brain.
• Real-world pay-off: The construct underlies diagnostic tools for dementia and helped seed knowledge-graph technology used in artificial intelligence.
• Predictive power: The spreading-activation model correctly predicts graded, distance-dependent priming, not just whether two words are related but roughly by how much.

Limitations

Several open questions temper these strengths.

• Model flexibility: Spreading-activation models have so many free parameters, such as arc lengths and decay rates, that they can fit almost any result, which weakens how testable they are.
• Where the store actually is: Even researchers who accept the hub-and-spoke account (see below) disagree about the balance between the amodal hub and the modality-specific spokes, and some deny a purely amodal hub exists at all.
• Overlap with episodic memory: Semantic knowledge is abstracted from personal episodes, so the two systems are not as cleanly separable as the theory implies; their retrieval processes can interact rather than compete.
• Narrow evidence base: Much of the classic evidence uses concrete, Western categories like birds and furniture, so it is unclear how well the models generalize to abstract or social concepts.

Contemporary Research

Recent work has tested and extended the neural account of semantic memory described above.

Aim. Rice, Lambon Ralph, and Hoffman (2015) set out to establish whether the brain’s semantic “hub” sits mainly in the left hemisphere, as some verbal-memory theories assumed, or in both hemispheres.

Method. They pooled 97 functional neuroimaging studies of conceptual knowledge in a meta-analysis, comparing how often each study found left-sided versus right-sided activation in the anterior temporal lobes.

Results. Activation was mostly bilateral and highly overlapping across almost every stimulus type, with only subtle leftward shifts for written words and word retrieval.

Conclusion. The semantic hub is functionally unified. It sits in both hemispheres, with local specialisation arising from graded connections to different types of sensory and motor knowledge. This reconciled the “amodal hub” idea with the distributed evidence, and became a linchpin of the modern synthesis.

Lambon Ralph, Jefferies, Patterson, and Rogers (2017) went further. They argued that semantic memory needs two systems working together. A representation system stores concepts, and a control system retrieves the parts that fit the current context.

Neither system works alone.

This explains why damage to the hub produces a uniform loss of concepts. Damage to the control network, in contrast, produces inconsistent, cue-dependent failures. The framework also grounds, computationally, the automatic-versus-strategic split Neely (1977) had already isolated behaviourally (see Semantic Priming above).

References

Arbib, M. A. (Ed.). (2002). Semantic networks. In The Handbook of Brain Theory and Neural Networks (2nd ed.) . Cambridge, MA: MIT Press.

Barr, A., & Feigenbaum, E. A. (1982). The handbook of artificial intelligence. Lost Altos, CA: William Kaufman.

Binder, J. R., & Desai, R. H. (2011). The neurobiology of semantic memory. Trends in cognitive sciences, 15(11), 527–536.

Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82(6), 407–428. https://doi.org/10.1037/0033-295X.82.6.407

Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of Verbal Learning and Verbal Behavior, 8(2), 240–247. https://doi.org/10.1016/S0022-5371(69)80069-1

Collins, A. M., & Quillian, M. R. (1972). How to make a language user. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 309-351). New York: Academic Press.

Das, J. P. (2014). Verbal Conditioning and Behavior. Oxford: Pergamon Press, Ltd.

Fishman, J. (2014). Advances in the Study of Societal Multilingualism. The Hague: Mouton Publishers.

Hodges, J. R., Patterson, K., Oxbury, S., & Funnell, E. (1992). Semantic dementia: Progressive fluent aphasia with temporal lobe atrophy. Brain, 115(6), 1783–1806. https://doi.org/10.1093/brain/115.6.1783

Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobiographical memory and perceptual learning. Journal of Experimental Psychology: General, 110 (3), 306

Jakobovits, L. A. (1966). Utilization of semantic satiation in stuttering: A theoretical analysis. Journal of Speech & Hearing Disorders, 31(2), 105–114.

Jakobovits, L. (1966). Semantic Satiation and Cognitive Dynamics (PDF). U.S. Department of Education.

Klein, S. B. (2013). Episodic Memory and Autonoetic Awareness. Frontiers in Behavioral Neuroscience, 7 (3), 1–12.
Jakobovits, L. A. (1962). Effects of repeated stimulation on cognitive aspects of behavior: Some experiments on the phenomenon of semantic satiation (Doctoral dissertation, McGill University Libraries).

McRae, K., & Jones, M. (2013). Semantic Memory. In Reisberg, Daniel (ed.). The Oxford Handbook of Cognitive Psychology. New York, NY: Oxford University Press.

Lambon Ralph, M. A., Jefferies, E., Patterson, K., & Rogers, T. T. (2017). The neural and computational bases of semantic cognition. Nature Reviews Neuroscience, 18(1), 42–55. https://doi.org/10.1038/nrn.2016.150

Meyer, D. E., & Schvaneveldt, R. W. (1971). Facilitation in recognizing pairs of words: Evidence of a dependence between retrieval operations. Journal of Experimental Psychology, 90(2), 227–234. https://doi.org/10.1037/h0031564

Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106(3), 226–254. https://doi.org/10.1037/0096-3445.106.3.226

Hoifung, P., & Domingos, P. (2009). Unsupervised semantic parsing. Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing.

Patterson, K., Nestor, P. J., & Rogers, T. T. (2007). Where do you know what you know? The representation of semantic knowledge in the human brain. Nature Reviews Neuroscience, 8(12), 976–987. https://doi.org/10.1038/nrn2277

Reiff, R., & Scheerer, M. (1959). Memory and Hypnotic Age Regression: Developmental Aspects of Cognitive Function Explored Through Hypnosis. New York, NY: International Universities Press.

Rice, G. E., Lambon Ralph, M. A., & Hoffman, P. (2015). The roles of left versus right anterior temporal lobes in conceptual knowledge: An ALE meta-analysis of 97 functional neuroimaging studies. Cerebral Cortex, 25(11), 4374–4391. https://doi.org/10.1093/cercor/bhv024

Sowa, J. (1992). Semantic networks. In Shapiro, S. (Ed.) Encyclopedia of Artificial Intelligence ., 2nd edition, Wiley.

Squire, L (1992). Declarative and Nondeclarative Memory: Multiple Brain Systems Supporting Learning and Memory. Journal of Cognitive Neuroscience, 4 (3), 232–243.

Sussna, M. (1993, December). Word sense disambiguation for free-text indexing using a massive semantic network. In Proceedings of the second international conference on Information and knowledge management (pp. 67-74).

Thompson-Schill, S.L. (2003). Neuroimaging studies of semantic memory: inferring “how” from “where” . Neuropsychologia, 41 (3): 280–292.

Tian, X., & Huber, D. E. (2010). Testing an associative account of semantic satiation. Cognitive Psychology, 60 (4), 267–290.

Tulving, E. (1972). Episodic and semantic memory. Organization of memory, 1, 381-403.

Tulving, E. (1984). Precis of elements of episodic memory. Behavioral and Brain Sciences, 7 (2), 223-238.

Vargha-Khadem, F., Gadian, D. G., Watkins, K. E., Connelly, A., Van Paesschen, W., & Mishkin, M. (1997). Differential effects of early hippocampal pathology on episodic and semantic memory. Science, 277 (5324), 376-380.

Warrington, E. K., & Shallice, T. (1984). Category specific semantic impairments. Brain, 107(3), 829–854. https://doi.org/10.1093/brain/107.3.829

Yee, E., Chrysikou, E. G., & Thompson-Schill, S. L. (2013). Semantic Memory. The Oxford Handbook of Cognitive Neuroscience. New York, NY: Oxford UP.

Further Information

• Tulving, E. (1972). Episodic and semantic memory. Organization of memory, 1, 381-403.