Basic Concepts About Concept Formation: Theory + Key Terms

How well do you know what a pizza is? Round, saucy, cheesy, comes in a box, served in triangles, right? Now do you know what Justice is? ……. Pizza is an easy concept to digest, pun intended. But Justice isn’t. Why? Intuitively speaking, a concept is formed when we learn examples and a definition, and acquire its mental representation by comparing and contrasting it with other concepts. A car is a concept. So is a monkey. Justice is too, but it’s a lot harder to represent. Even brands and love are concepts (but not really). There are many incomplete theories of concept formation and different ways to evaluate a concept. Here’s a detailed overview of how we form concepts, how we process known concepts, and how we can maximize the capacity to learn new concepts. Skip to the end if you just want to know how to use the theory to improve a learner’s conceptual understanding.

How are concepts formed?

Definition: A concept is defined as a meaningful unit of knowledge that represents a real-world object or a cohesive entity with properties and is used to interpret new information and make sense of existing information. The source of this information is sensory which is translated as a perception (readable by the mind) and connected to (and influenced by) existing knowledge which is stored as a memory. The processing is done via cognition and metacognition.

The classical view is that we learn a concept’s defining features and its examples. For example, a child learns that living things move and non-living things don’t move, so things that move belong to the concept of “living.” Then the concept is updated with more criteria. A child then learns that something that used to move and has stopped moving would belong to a new concept called “dead.” So both concepts create a new concept. This view is limiting because a few defining criteria don’t account for natural variations and diversity. Pure definitions can’t exist for something like “greek food,” but can exist for “pizza”. Yet, both are equally valid concepts.

The probabilistic view is that we learn concepts by recognizing the most common features among many examples. Co-occurring features are averaged out and stitched together to form a category. A dog has co-occurring features such as 4 legs, barks, wags tail, and comes close to humans. We group these features in a process called “feature clustering” or “chunking” to simplify them as one single thing instead of many things. These features are averaged across many experiences with dogs that develop a rigid concept of a dog. These are then updated with new criteria with new rules such as a digital dog, which doesn’t come close to humans but looks exactly like one on a screen. The probabilistic view is more feature- and property-driven without making them hard criteria or definitions. As long as they co-occur with reasonable flexibility, they become a unified concept. These concepts also develop by learning many new examples and finding what’s common in them all. Looking at a number of cats, one can create loose properties for size, but fairly stable properties for facial features and movement, all of which co-occur among most examples. However, this view fails to account for diversity where a single property is more important than all other properties. Dinosaurs, for example, are defined as creatures from a specific era, but the mental image has nothing to do with that era – it could be a pterodactyl (flying dino), t-rex (land dino), or a spinosaurus (water dino, unless it has an identity crisis, yet again). All of them are different forms with very few common features.

The prototype view looks at how each concept has a prototype – a typical example that represents a category. Here, the prototype would have approximate features and new examples are compared with the prototype before judging them as belonging or not belonging to that category. For example, a german shepherd can be a prototype and all dog-like creatures are judged on how similar they are to the prototype. If they are very similar, the example is assigned to the concept of a dog. If they are remotely similar, they are not. This approach may function on some level but doesn’t explain how wolves are not dogs.

The abilities view takes a different but meaningful approach to concept formation that the prototype view fails at. Concepts are essentially defined by how they can stay separated from each other. The concept of a cat and dog can be well separated, so there are 2 emerging concepts. But the concepts of trust and reliability are not easily differentiated, so they might as well be one. Here, the core of concept formation is the ability to differentiate one cluster of information from another.

The exemplar view is a combination of the ability view, prototype view, probabilistic view, and classical view where examples are the foundation of concept formation. Ideal examples, sometimes called “referents” or “exemplars” become the central focus of a concept, and not the concept itself. New examples modify the concept and more examples differentiate a concept from other ones. The variety of properties that exist among all examples of a category create the concept as an “abstraction” of the “exemplar”. Imagine a dinosaur, perhaps you thought of a T-rex. But there are more dinosaurs. Here, all dinosaurs need not have all properties, but a few properties like size, era, teeth, and walking style are enough to recognize the abstract idea of a dinosaur which is only reasonably understood based on the examples. However, even a tiny creature, which more resembles an insect today, can be identified as a dinosaur based on the number of examples one has that feed the concept of a dinosaur. This view also allows a typical example to represent a concept instead of the concept being a typical thing that represents all of its examples. So, for example, T-rex represents dinosaurs. But dinosaurs don’t represent T-rex when we think about “dinosaurs”. However, this view is also not enough because it fails to explain how we understand love or more abstract ideas such as squareness.

These ideas have an interplay with language and sensory features with updates coming from previous knowledge. So I’ll introduce a new view called the dynamic-meaning view. It addresses a concept and its interplay with abstract thought, sensations, interaction with the world, and language use^[1]. A concept is formed in an existing knowledge network of properties and representations that are connected with many mental representations of concepts, mental representations of examples, real-world structure and function of a concept put in words, and contexts. We first understand concepts^[2] by clubbing features/properties of examples and categories (size, weight, function). We then place them in a classifying system for categories (animals, vehicles, food). We then differentiate and form unique concepts by looking at rules, similarities, and differences^[3] between multiple categories (animals vs. trees) and objects/examples (car vs. plane) simultaneously.

This process lets us extract the most minimalistic features of concepts as abstract ideas and store them in memory. These are base-level sensory attributes called “perceptual properties” that have very little independent meaning. Properties like length, movement in a specific pattern, textures, visual angles, location, association with something, emotions related to a situation, new sounds, etc., and contextual variables like what co-occurs, the purpose of analysis, the typical environment where a concept is noticed or expected, etc., are included in perceptual properties. Most of these perceptual properties are extracted from a concept since it’s not yet evident if a baby is born with inherent concepts. However, babies and animals can identify minimalistic perceptual properties. These perceptual properties are inherently attached to their original examples (roughness of skin, shape of a human), but their representation can be detached to and identified in new concepts. This is where language and experience create a way to identify those perceptual properties and interpret new concepts (latex feels different from skin, a shadow looks like a human). Then the brain creates 2 additional properties which are independent of the concept: A clear mental model of the perceptual properties (shape, taste, feelings something evoked) and a language-based label (words as descriptors).

Categorizing 2 different things with labels gives them unique meaning, just because of the categorization^[4] with labels. These additional properties then become an abstract way to separate 2 concepts. That means, we tend to give this classification additional meaning because of the label. The label represents more than the thing that is labeled because of clearly stored perceptual features and words. Eventually, new concepts are represented as metaphors and analogies for those perceptual properties, and labels that become their own models. As the total list of perceptual properties and models increases, they become the collective pool of knowledge. This is aligned with gestalt psychology and embodied cognition views of concept formation.

The more details there are in a person’s conceptual understanding, the more they become an expert. Concepts are updated with new information as one learns more nuances. And more nuances allow a person to re-organize concepts.

Concept formation is inherently a noisy classification task. Each perceptual input – sensory and descriptive – allows a person to classify a concept and differentiate it. Noise indicates irrelevant or confusing information that makes classification and recognition difficult. Attempts to find patterns for concepts is essentially a noise-reduction mechanism. Grouping stars and separating out the planets from the night sky is a noise-reduction method to conceptualize the night sky.

Concepts about concepts

Concepts are organized as knowledge and have 7 primary properties: Concept hierarchies, Concept associations, Concept clusters, Concept categories, Concept exemplars, Conceptual criteria, and Concept features.

1. Conceptual hierarchies: These are vertically stacked concepts ranging from the broadest category (superordinate level) to the most specific category (exemplar level or subordinate level). Furniture is a broad superordinate level containing chairs, tables, and sofas, which are sub-categories or the subordinate level.
2. Concept associations: Each concept has interrelated concepts. The concept of male is associated with a female, and the concept of trees is associated with plants and fruits. Associations occur in 3 dimensions: Properties/features, sub-categories and parent categories, and related concepts. These form one’s knowledge base.
3. Concept clusters: Concepts are grouped together based on similarity. Furniture, cutlery, tools, etc., are concept clusters under the larger category of home utilities.
4. Concept categories: Concepts are represented by words (labels) or informational patterns (models). These contain collections of examples with common features. These are also called conceptual classes. “Dogs” is a concept category; so are “mountains.”
5. Concept exemplars: A concept has ideal or model examples that are observed and classified into categories that are typical/prototypical for that category. Specific breeds of dogs or individual pet dogs are exemplars. Exemplars are ideal examples and examples are specific objects that belong to a category. A digital pet dog, may not be an exemplar but it is an example.
6. Conceptual criteria: A concept has a definition that has certain criteria. A square has 4 sides of equal lengths with right angles between adjacent sides. Not all concepts are easily defined, particularly the ones that are naturally found and we have to reverse-engineer to discover the concepts. The strictness of criteria that define a concept can make a concept meaningful or not meaningful.
7. Concept features/properties: Each concept has perceptual features that our senses process and the brain encodes before it becomes a memory, which then becomes a mental image. These features or properties are the rough definitions that describe the structure and function of a concept. A hammer has features such as a flat heavy head and a handle that is used to apply force onto something else, for example. The precision and volume of properties that belong to a concept in one’s head describe how well one understands it.

Subordinate vs. Basic vs. Superordinate categories of concepts

A basic concept is typically a concept that is the easiest to think of, most relevant, and perceived in a context. For example in a dining context, a fork or spoon is more basic than cutlery. This is because, in everyday use of a concept, there is a typical use-case and perceptual level that is most meaningful on average (its construal). Saying, “give me cutlery” isn’t nearly as useful as saying “give me a fork”. The brain inherently prefers concreteness in most cases with a fair balance of usefulness and generalizability. For example, saying “give me a tuning fork or a skewer” while eating would be less useful and less general, even though it is more concrete.

With a basic concept like monkeys, it has a global, parent category above it called the superordinate level or supraordinate level (mammals) and a specific example or a finer category below it called the subordinate level or the exemplar level (bonobo monkeys). These form a concept hierarchy.

Superordinate concept:
Mammals, Furniture, Forests

Basic concept:
Mammals (Cows, Monkeys, Dogs)
Furniture (Table, Chair, Cupboard)
Forests (Evergreen forest, Rainforest, Deciduous forest)

Subordinate concept:
Mammals: cows (Black Angus, Red Angus, Hereford), monkeys (Bonobo, Macaque, Orangutans), dogs (Shih Tzu, German shepherd, Chihuahua)
Furniture: table (Dining table, Study desk, Conference table), chair (Beanbag, Stool, Office chair), cupboard (Shelf, Drawer stacks, Modular kitchen)
Forests: evergreen forest (Mahogany, Ebony, Rosewood), rainforest forest (Kapok, Rubber, Xate), deciduous forests (Oaks, Beeches, Chestnuts)

In the entire hierarchy of concepts, the “type” or “sub-types” or “examples” or “sub-categories” of a concept are usually basic concepts or subordinate concepts belonging to a superordinate concept.

Concepts organized as superordinate, basic, and subordinate mimic how sets are defined in mathematics – a group of elements that belong together. The most meaningful superordinate level would be everything contained in the universe and not contained in it. The most meaningful subordinate level would be the smallest stable physical entity, like quarks that create atoms.

Properties of a concept

A single concept has many properties or conceptual features. A bird has properties of flying, being light, moving fast, having a beak, etc. A specific example of a bird will have additional properties. Here, a sparrow will be a nested concept within the category of bird. These properties, however, are not straightforward. Not all properties of a concept^[5] are activated in the brain when a concept is activated. The properties that are highlighted are context-driven in some cases and fairly universal in other cases.

1. Context-independent properties: These are features or properties of a concept, category, or exemplar that are relatively always recognizable. For example, flying is a dominant property of birds in most cases.
2. Context-dependent properties: These are features or properties of a concept, category, or exemplar that are relatively meaningful only in specific contexts. For example, being bird-boned or calling someone a tiny bird in the case of physical immunity or robustness is context-dependent and wouldn’t make sense while talking about different species of birds.

Contexts also evoke certain properties and trigger the activation of a specific example more readily than other examples. That is, context can bias us to invoke a specific concept. Saying a bird walked across the barn is more likely to invoke the concept of a chicken instead of an eagle. Even though both concepts are nested within the concept of a bird, which is technically the concept referred to in the example (pun not intended). Similarly, saying a biker went up a rocky terrain would imply the category of bike refers to a sub-category of a mountain bike and not a racing bike.

Concept vs. Construct

Concepts are generally considered concrete ideas like a car, animal, tools, etc., and constructs are abstract ideas like love, evenness, roughness, morality, etc. According to most theories, concepts and constructs are a continuum where concepts are connected to real-world objects that can be sensed, and constructs are connected to derived meanings from those real-world objects. Water is a concept and wetness is a construct. A brand is a construct with conceptual associations that are tangible. Like Zomato and Ubereats is a construct associated with food and comfort. Constructs become more symbolic than sensory.

Abstractness vs. Concreteness

Concreteness describes the precise features of a concept that don’t change based on different people’s perspectives. For example, most people will agree the concept of a beach is a sandy area connecting to an ocean. Abstractness describes the essence of a concept without much detail. Like representing a building as a basic tall cuboid or metaphors like a beach’s sandy-ness is like your personality – with every wave, something changes but still, you remain the same. Similarly, lust is a real-world phenomenon connected to mating and love is an aggregate feeling derived from many interactions in the real world. Generally, abstract constructs are more difficult to process^[6] than concrete concepts, so the brain readily thinks in concrete terms.

Abstract features of a concept occur in 2 forms:

1. The common denominator and shared feature: Comparing 2 objects like sand and tree barks with their “roughness.” This also includes long jumps in associations like saying “red is a social color”. Here, red’s arousal and attention-grabbing loud nature can be compared to a noisy social context of high arousal and attraction between people.
2. The smallest element capturing the most of a concept: Representing objects with their basic shapes without details or saying lukewarm water is soft and sparkling water is sharp.

The kiki-bouba experiment shows how humans abstract and find cohesive patterns between multiple ideas, objects, sensations, and perceptions. In the experiment, 90% of people always assign the word “bouba” to a round shape and “kiki” to a starry shape. The explanation is that the word “bouba” and the shape have an inherent roundness that makes them belong together and the word “kiki” and the starry shape have an inherent sharpness that makes them belong together.

Construal levels of details in a concept

A construal level is a level at which people perceive a concept typically. A high construal is processing a concept at a vague, abstract, or global level. For example, processing a phone as a communication device (the superordinate level) is a high construal. A low construal is processing a concept at a detailed, specific, or local level. For example, processing a phone as a Samsung Note 10 or iPhone 13 (subordinate level).

A person will “construe” or comprehend a concept with a specific set of features, associations, and depth of processing. An engineer would think of a car in different ways than a lawyer would. Similarly, you may think of people from your past with a set of characteristics that were typical during that time. From all the possible features of a concept you can think of, a subset is used to represent the concept in a given context. So effectively, each object and idea has its own-default construal – the set of attributes we readily perceive. We can then change the construal to a different set of attributes. For example, the concept of “parents” can have construal levels such as “my own parents when they became parents,” “my own parents right now,” “2 people who raise and nurture a child,” etc. The construal level depends on the associations (how they nurture), conceptual hierarchy (relation to children), or concreteness (their behavior and looks vs. their purpose in life) that you are focusing on.

A construal level essentially describes 2 types of information about a concept:

1. Its position in a hierarchy
2. Its associations and typical features for an observer

The construal level of a concept you invoke in conversation affects communication too. Saying you want to Uber home can refer to the low-construal “Uber” or the high-construal “Transport”. Most miscommunication comes from a mismatch of construal levels of speaking.

The superordinate level, high construal level, abstractness, and constructs become one and the same. Justice, for example. Similarly, the subordinate level/exemplar level, low construal level, concreteness, and concepts become one and the same. Pepperoni pizza, for example.

Prototypical features vs. Stereotypical features

A stereotype is a template of a concept that has the most typical features clubbed together or is the best example to represent a category (the superordinate category). A stereotypical metalhead (fan of rock music) would have long hair, wears a band tee, and uses the metal horn hand sign. A stereotypical actor would be physically attractive and appears on screen in movies and TV shows.

A prototypical concept is a concept with minimum criteria to become a concept or represent a category. A prototypical dog would have 4 legs, a tail, a snout, and barks. A prototypical car has 4 wheels, 1 driver seat, 3 passenger seats, and a roof. Before finalizing a finished product, companies build a prototype that approximates the final product.

In many ways, a prototype and stereotype are the same because they are the best examples of a category, but stereotypes are used more in a social sense with negative connotations, and prototypes are used in a non-social context.

The famous “duck test” is a heuristic to identify a concept. If it looks like a duck, swims like a duck, and quacks like a duck, it’s probably a duck. This heuristic (decision-making shortcut) refers to a duck’s prototype.

Spreading activation and associative network

Concept categories and concept features are situated in a virtual map of nodes that carry information about the categories, properties, and examples. The nodes are connected in a network. Spreading activation is a method of activating a node. When attention is given to a node, its information is highlighted, and related information is partially highlighted. Activating a node of “trees” will activate nearby nodes that contain “plants,” “forests,” “fruits,” etc. Spreading activation describes how attention activates concept clusters or feature clusters at a particular construal level. High-level concepts like broad categories (electronics, clothes) activate its examples and other high-level concepts such as e-commerce and shopping mall. Low-level concepts like specific categories or exemplars activate other exemplars – activating apples brings grapes, and banana’s into awareness but is less likely to bring herbs into awareness (which may be further away in the associative network).

When is a concept meaningful?

Concepts are inherently meaningful when they are differentiated from each other but still maintain some coherence in a hierarchy AND context. The concept of 5 atoms together is not very meaningful unless it’s important in a physics experiment. Concepts become meaningful when they have 4 properties:

1. They are stable: A car looks like most cars. The stability is derived from a set of features and patterns among many examples or observations. You may observe the sun’s rising and falling patterns and understand the concept of a day. Perceptual stability is a minimum requirement for meaning. This comes with definitions, criteria, models, and sensory input.
2. They have an inherent pattern: A pattern emerges when an observer compresses information in a way that’s easy to process. The goal of finding a pattern is to reduce noise in what we perceive and make it coherent. The reason for this compression is ease of thinking and a biological preference for perceptual fluency (reduced difficulty in making sense of information because effort is biologically expensive). This happens when features are clustered together in a way they get a new identity. Hundreds of dots arranged around a central point can be hard to process if analyzed one by one. But taken together, those dots form a circle – a compressed version of those dots. In short, the properties of a concept are inherently cohesive and often co-occur.
3. They are differentiated: The pattern established between observations can be separated from the pattern established with another set of observations. The concept of a chair can be separated from the concept of a table. One concept can also be used to define another concept. Noise can be defined as something that is NOT a pattern. A car can be defined as a vehicle that isn’t a bicycle, a motorbike, a truck, or a bus.
4. They are useful in building knowledge: The concept has some value to humans and can help to process new information. One concept can be used to describe another one. For example, an airplane can be defined as a flying bus. And flying can be defined as the way birds move in the sky. A bird can be defined as a rat that moves above the ground. And so forth. Each of these definitions can then be refined and developed into a meaningful concept.

Loose conceptual boundaries (differentiation) or insufficient criteria (low stability/weak patterns) can create errors in understanding and recognizing a concept. Error’s in concept recognition – correctly classifying a chicken as a bird or a dinosaur or food depends on 2 properties that are contextually important: Specificity and Sensitivity.

Specificity and Sensitivity in concept recognition

Specificity (true negatives): When you do not assign an example to a category it should not belong to, your conceptual understanding is 100% specific.

Sensitivity (true positives): When you correctly assign an example to a category, you are 100% sensitive to understanding the concept.

If you correctly differentiate between a car and a bike 100% of the time, your conceptual understanding is 100% specific and 100% sensitive because you pass 4 judgments: A car is not a bike and a bike is not a car (specificity), and a car belongs to a car concept and a bike belongs to a bike concept (sensitivity).

If you aren’t specific or sensitive, it means one of 5 things.

1. Your conceptual understanding is weak
2. The concept itself is poorly defined
3. The concept requires contextual information
4. You are defining a concept at a high construal level or invoking a very high superordinate category
5. You discovered a new pattern between the example and category

Sensitivity and specificity are important to becoming an expert in an area. For all concepts important in a domain, A=/=B, A=/=C, A=/=D, etc., is as important as saying A is A. For example, a doctor would be able to differentiate many diseases and use tools to correctly identify them. They identify symptoms as a cold, covid, or allergies. Sometimes, there is a misdiagnosis where they fail to identify an allergy as an allergy. Their expertise depends a lot on the correct identification of a problem and choosing a treatment that belongs to that disease.

Both properties rely on context-dependent concept activation and are severely affected by the amount of noise in one’s learning. Noise is ambiguity in the information that doesn’t help with concept recognition, concept formation, and concept differentiation. The brain has neural processes that reduce the impact of noise on perception. Background information, which we call context, allows us to interpret features or pattern ambiguity into something meaningful.

How to help learners learn new concepts using concept formation theory

1. Give lots of examples of things that belong to a category and things that do not belong to it
2. Identify co-occurring features between 2 related or unrelated concepts
3. Identify exceptions or improperly classified concepts
4. Refer to examples as examples and concepts as concepts, instead of interchanging the two in daily conversation
5. Assign visual models, sounds, movements, touch, taste, smell, and context when possible
6. Label concepts with words
7. Have variety in examples and properties
8. Differentiate between similar concepts
9. Draw similarities between 2 concepts to see if they converge or don’t
10. Represent an associative network of concepts
11. Discuss concepts at their smallest feature levels and highest abstract levels (construal levels)

Do leave a comment asking for explanations or to request a deeper article about a particular concept!

Sources