Chapter 8: Concepts, Categories & Knowledge Organization

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement, not replace the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

Have you ever found yourself wrestling with a new job title?

Or maybe you're trying to organize a file folder on your computer and you realize you have to make a tough call.

Like, does this new thing fit here, or does it belong somewhere else entirely?

Oh, all the time.

Or you're looking at a college course catalog, trying to figure out if psychology 101 counts for your social science or your natural science credit.

Exactly.

Or a more high stakes version, you're at the doctor's office waiting for a diagnosis.

Is this pattern of symptoms just the flu,

or is it something much more serious?

Well, those moments, that pressure to classify things, they're actually perfect examples of one of the most fundamental things our brains do every single day, sorting the chaos of the world into, well, manageable boxes.

And that sorting process is the central focus of this deep dive.

Today, we are cracking open the cognitive machinery behind concepts and categorization.

And this isn't just some abstract topic.

This is the scaffolding of how we think.

If you don't get this, it's really hard to understand how the mind builds knowledge or solves problems.

Totally.

Our mission here is to dive into the core building blocks of human thought, looking at chapter 8 from the cognitive psychology canon.

We're examining how our minds take this just overwhelming diversity of experience.

All the different dogs you see, every car on the road.

All of it, and create order and predictability out of it.

So before we jump in, let's just nail down the two key terms, because the language here really matters.

Good idea.

When we say a concept, we're talking about the mental representation.

It's that idea in your head.

Your concept of a dog is like four legs, fur, barks.

It's the abstract summary.

And the category is the actual group of things out there in the world.

It's the box you put things in.

So when you decide your psych course is a social science, that's the category.

And your idea of what a social science course is, that's the concept that lets you make that decision.

The benefit here is just immense.

I mean, without concepts,

mental life would be pure chaos.

You'd be overwhelmed.

You'd have to treat every single chair, every single tree as a totally unique thing.

You couldn't learn from experience.

So concepts, let us take something brand new, a dog breed you've never seen before, and treat it like other things we already know.

It lets you predict things, and it dictates how you should act.

If I see a four -legged animal running at me and classify it as a dog instead of a wolf, my next move is going to be very different.

That instant decision is what this is all about.

OK, so let's unpack the theories here, because this field really went through a revolution back in the 70s.

We have to start with the old -school dominant theory.

That's right.

We start with what's called the classical view.

And it was the main way of thinking about this stuff for, well, centuries.

It goes all the way back to Aristotle.

And the core idea of this classical view is it's very neat and tidy.

It says concepts are defined by a list of necessary and sufficient features.

Right.

It's totally black and white.

Let's break that down, because both of those words are critical.

OK, so first, the features have to be individually necessary.

That means every single example of the concept must have that feature.

No exceptions.

Like a triangle.

It must have three sides.

If it has four, it's out, disqualified.

Exactly.

And second, the features have to be collectively sufficient.

So if something has all the features on the list, it is automatically 100 % an instance of that concept.

So three -sided, closed, and geometric figure.

If it checks all those boxes, it's a triangle.

Period.

And this view has some pretty big implications.

It means concepts are just abstract feature lists, not specific examples.

It means category membership is all or none.

And this is a big one.

It means all members of a category are created equal.

There's no such thing as a better triangle or a worse triangle.

It's a very logical, very clean way to think about it.

The problem is it just doesn't seem to be how human minds actually work.

Right, this is where the whole thing just gets smashed to pieces by the work of Eleanor Rauch in the 1970s.

Her work was a bombshell.

She introduced this idea of typicality.

She found that people consistently judge some members of a category as being better or more representative than others.

Which totally contradicts the classical view that all members are equal.

Completely.

It's that classic bird example you always hear about.

If I ask you for a bird, you'll probably say.

A robin or a sparrow.

Exactly.

A very good example.

But what about a chicken or a penguin?

Not so good.

Less typical.

Even though a penguin has all the necessary features of a bird, feathers, beak, lays eggs, it just doesn't feel as birdy as a robin.

And our brains actually process them differently.

We have hard data on this.

Right, the sentence verification tasks.

People are way faster to say yes, that's true too.

A robin is a bird than they are too.

A chicken is a bird.

That speed difference shows there's a cognitive cost.

It takes more effort to connect the atypical example to the category.

And if you ask people to just list instances of a category like furniture, they'll say chair and sofa first.

They won't say bookends until much, much later, if at all.

Then there's semantic priming.

If you flash the word robin on a screen, people are faster to recognize the word bird right after.

But if you flash penguin first, the priming effect is much weaker.

All this evidence just piles up against the classical idea of a strict feature list.

And it's not just about typicality.

The classical view also fails on the idea of fuzzy boundaries.

McCloskey and Glicksburg did this great study in 78.

They asked people about atypical items, like are bookends furniture?

And what did they find?

A lot of disagreement.

But here's the killer finding.

People weren't even consistent with themselves.

They'd come back a week later and give a different answer.

Wow, so I might say yes on Monday and no on Thursday.

Exactly.

And that just destroys the idea of a clear fixed definition in your head.

If the rule was there, your answer should be the same every time.

And the final nail in the coffin is just, you can't do it.

Try to define game with necessary and sufficient features.

People have tried for centuries, it's impossible.

You just can't, which means we needed a whole new way of thinking about this.

Okay, so if the mind doesn't use a strict list of rules, what does it use?

This brings us to the first major alternative,

the prototype view.

Right, and this view shifts from using necessary features to using characteristic features, features that are typical but not required.

So instead of a rigid definition, a concept is a mental prototype.

It's like an idealized average or a summary of the category.

The key here is that no single feature absolutely has to be there.

An object is categorized based on how many of these characteristic features it has.

The more it has, the better a match it is to the prototype.

This reminds me of Wittgenstein's idea of family resemblance.

It's the perfect analogy.

Think of a family like the Smith brothers in the textbook figure.

They might all share traits like light hair, a mustache, big ears.

But not every brother has every single trait.

Exactly, one has the hair and ears, another has the ears and mustache.

But the brother who has the most of those shared features, he's the most typical looking Smith brother.

He's the prototype.

The category holds together because of this web of overlapping features, not one single feature they all share.

And Rosh and Mervis actually tested this.

They had people list attributes for things like chair and clock, which are both in the category of furniture.

What did they find?

The prototypical items like chair and sofa shared way more features with the general idea of furniture than the atypical items like clock.

And crucially, almost no features were true for every single item in the category.

So that's why a robin feels birdier than an ostrich.

Right, the robin just checks more of the boxes on the characteristic feature list for our bird prototype.

It's small, it flies, it sings.

The ostrich misses a bunch of those.

This also leads to another big idea from Rosh, the basic level of categorization.

Yeah, this is so important.

The idea is that our categorization system has to balance two competing goals.

Okay, what are they?

First, we wanna group similar things together to be informative.

And second, we want to keep our categories distinct from each other to be efficient, or what we call cognitive economy.

So the basic level is the sweet spot.

It's the sweet spot.

Think of piano or guitar.

The things in that category are really similar to each other.

A grand piano and an upright piano are pretty similar.

Right, and at the same time, a piano is very different from another basic level category, like a trumpet.

It gives you the most information for the least effort.

As opposed to the superordinate level, like musical instrument, which is just too broad.

A piano and a guitar don't share that much.

And the subordinate level, like grand piano, is too specific.

It doesn't give you much more useful information than just piano.

So we naturally gravitate to that basic level, dog, car, apple.

But the prototype view isn't perfect either, right?

It has its own set of problems.

For sure.

One big one is that it has trouble with conceptual boundaries.

What do you mean?

Well, think about a tiny Pomeranian and a huge Great Dane.

They're both dogs.

But perceptually, that Pomeranian might look a lot more like a cat than it does a Great Dane.

That's true.

A simple prototype model based on averaging features struggles to explain why we draw such a hard line between dog and cat in that case.

So the boundaries seem to come from something other than just raw similarity.

Right.

They come from our knowledge about the world, about biology.

We learn which features tend to cluster together.

Another problem is context dependence.

Yes.

Typicality changes depending on the situation.

A robin is a typical bird in your backyard.

But if I ask you about birds in a barnyard.

Then suddenly a chicken becomes much more typical.

The prototype isn't stable.

It shifts with the context.

And then there's the killer critique, the odd number problem.

Ah, yes.

So researchers found that people will happily rate the number three as a better or more typical odd number than, say, 57.

Wait, that makes no sense.

Odd number has a perfect classical definition.

An integer not divisible by two.

There shouldn't be any typicality.

Exactly, that was the point.

The fact that people still give typicality ratings for a perfectly defined concept suggests that maybe the rating task itself is just a weird thing people do.

So it might not be reflecting the true structure of the concept in their head.

It might just be reflecting things like frequency or simplicity.

This was a huge challenge and it opened the door for a totally different idea.

Which is.

Maybe we don't store an abstract summary at all.

Maybe we just store memories of actual examples.

This is the exemplar view.

So my concept of dog isn't some idealized average dog.

It's my memory of Fido and Rover and every other specific dog I've ever met.

Precisely, the concept is the collection of those individual instances or exemplars.

Okay, I can see how that would explain a lot.

You don't need to state necessary features because there aren't any, it's just the collection.

And it explains why we get stuck on things like the tomato.

Is it a fruit or a vegetable?

The exemplar view would say it's because a tomato is pretty similar to some of our fruit exemplars, like apples, but it's also similar to some of our vegetable exemplars, like squash.

It creates a conflict.

And typical things like robins are processed faster because we've seen more of them, so we have more exemplars stored.

Or they're just really similar to a lot of our other bird exemplars.

It's a really powerful model, but it has one massive glaring problem.

Let me guess.

Cognitive economy.

It sounds like you'd have to store a memory of every single thing you've ever seen.

That's the one.

It's incredibly unconstrained.

Which instances get stored?

How do you search through that gigantic mental library every time you need to classify something new?

It's a huge theoretical hurdle.

Okay, so this problem of storing just a massive amount of information, it kind of pushes us beyond just comparing features, right?

We need some kind of structure.

Exactly.

And that brings us to the schemata view.

Schemata are basically just organized frameworks for our knowledge.

So it's not just a bag of features or a pile of examples.

It has slots and variables, and it can be hierarchical.

Right, you could have a big super schema for college orientation, and inside that, you'd have a sub schema for meeting your roommate.

And that schema has slots for things like first conversation, rules for the room, and so on.

And this view is nice because it kind of combines the best of the other views.

It has abstracted information like the prototype view, but it can also store specific instances, like the exemplar view.

But it's not the final answer, is it?

Sounds a little fuzzy.

How do we know which schema to use?

How do they get updated?

That's the critique.

It's not always specific enough to be easily tested.

To really get at the why of categorization, we have to go one level deeper.

To the knowledge -based view.

Right, and this is where things get really interesting.

The KBV says that when we classify something, we're not just matching features.

We're using our knowledge and our theories about the world to justify why things belong together.

So the relationship between a concept and an example isn't just about similarity.

No, it's more like the relationship between a scientific theory and the data that supports it.

We're trying to explain why the group makes sense.

The classic example for this is the one from Barsalut.

Oh yeah, it's perfect.

Imagine this category.

Children, pets, photo albums, and cash.

Okay, physically they have nothing in common.

A dog doesn't look like a stack of bills.

Right, none of the similarity -based models could group them together.

But if I give you one piece of context,

your house is on fire.

Then they immediately become the category of things to save.

Exactly, and that category only makes sense because of your underlying knowledge, your theory, that these things are precious and irreplaceable.

The concept isn't a feature list, it's an explanation.

So this is what separates the different approaches.

You have the similarity -based ones.

That's your classical, prototype, and exemplar views.

They all rely on matching an item to some kind of standard.

And then you have the explanation -based ones.

Which includes schemata and this knowledge -based view.

They focus on the meaningful relationships between the items.

And this really gets at a fundamental problem with the whole idea of similarity, doesn't it?

The philosopher Nelson Goodman pointed this out.

He did.

He said similarity by itself is meaningless.

A fork and a spoon are similar because we use our knowledge of utensils to focus on the relevant features, their metal you eat with them, et cetera.

But take a plum and a lawnmower, are they similar?

Well, they both weigh less than 100 kilos.

And neither one can play the piccolo.

Exactly, you can find an infinite number of shared properties if you don't know which ones matter.

Similarity is useless until your knowledge tells you which features are the relevant ones.

So all of these models, they're all trying to find a balance, right?

Between cognitive economy and informativeness.

It's the essential trade -off.

You can't be too economical and have one category for things that's totally uninformative.

And you can't be too informative and have a unique category for every single object that would just overload your brain.

It's about finding that perfect middle ground.

Okay, so that's the structure of concepts, but how do we actually form them?

How do we learn them?

This takes us back into the lab.

And we have to start with the classic work from Bruner, Goodnow, and Austin in 1956.

They were studying how people learn nominal concepts.

Those are the concepts that actually do have clear, necessary, and sufficient features, the ones that fit the classical view.

Right, their setup was really clever.

They had these cards with geometric figures that varied on four dimensions, shape, color, number of shapes, and number of borders.

And the experimenter would have a rule in mind, like black circles.

The person's job was to figure out that rule by picking cards and getting feedback.

Yes, that's one, or no, that's not.

And they found people used a few different strategies to crack the code.

The first one was simultaneous scanning.

That sounds intense.

It is.

You try to hold all possible hypotheses in your head at once and eliminate them with each card you pick.

It's very efficient if you can do it, but it puts a massive load on your working memory.

I feel like my brain would just crash.

A lot of people's did.

So a more common one was successive scanning, which is much more manageable.

You just test one hypothesis at a time, is the rule black figures.

You test that until it's proven or disproven, then you move to the next one.

Slower, but easier.

Okay, and what was the third one?

The most effective one usually was conservative focusing.

You find one card that's a positive example, a focus card.

Then you pick new cards that only change one single thing.

So if your focus card is two black crosses, you might then pick two white crosses.

And if that's also a positive example, you know that color doesn't matter.

Exactly.

It's systematic, efficient, and pretty easy on your working memory.

It's actually a lot like how people play the game Mastermind.

So the big takeaway here is that when concepts are defined by clear rules, people become analytical.

They test hypotheses to find those rules.

But that strategy completely changes when the concept is fuzzy.

Which brings us to acquiring prototypes and the work of Posner and Keele with those dot patterns.

Right.

They wanted to see if people could form a prototype of something they never actually saw.

How'd they do that?

They created a prototype, like a perfect triangle made of dots, but they never showed it to the participants.

Instead, they only showed people distorted versions of it.

Some with low distortion, some with moderate distortion.

And the interesting part was the test.

They showed everyone brand new, highly distorted patterns, and the people who had learned from the moderate distortion examples were way better at classifying the new ones.

Why is that?

That seems counterintuitive.

Because they didn't just learn the central tendency, the prototype.

They also learned about the category's variability.

They got a better sense of how much an item could deviate and still belong.

So learning the boundaries is just as important as learning the center.

For fuzzy categories, absolutely.

It shows that the strategy we used to learn depends entirely on the kind of concept we're learning.

But there's a third way, too.

Right, which is implicit concept learning.

We don't always use these logical, analytical methods.

Sometimes we just absorb things.

Arthur Reber's experiments with artificial grammars are the classic example of this.

He had people memorize these long strings of letters.

And secretly, the strings were generated by a really complex set of rules, like a hidden grammar.

Right, and people who learned the grammatical strings got better at it than people learning random strings.

The structure helped them, even though they couldn't tell you what the rules were.

But here's the crazy part.

This is the best part.

He ran another condition where he told people there were complex rules and that their job was to figure them out.

And they did worse.

They did worse.

When you force people to look for rules in a complex system, they tend to invent wrong, overly simple rules that actually hurt their performance.

So sometimes it's better to just soak in the examples, the exemplars, and not try to analyze it.

That's Reber's conclusion.

Brooks called this non -analytic concept formation.

You're not testing hypotheses.

You're just storing representations of the individual example.

Like in his hieroglyphic study, people learned what the symbols meant without ever consciously realizing it.

They just used similarity to old examples to classify new ones.

It's a powerful but unconscious way of learning.

And Brooks suggested there are a bunch of real life reasons why we often rely on storing these exemplars.

OK, like what?

Well, first, we often need to tell individuals apart within a category.

Your family dog, Rover, and the scary attack dog down the street are both dogs.

But you really need to treat them differently.

Right, the prototype dog isn't enough information there.

Second, we see the same instances over and over in real life.

You see Rover every day.

That repetition reinforces the exemplar in memory.

Real world things vary on tons of complex dimensions, making rule finding really hard.

And any given thing belongs to multiple categories at once.

Rover is a dog, a pet, a source of mud.

You need the specific data from the exemplar to handle all that context.

And finally, we often learn things without knowing how we'll need to use that information later.

Storing specific exemplars gives you the most flexibility down the road.

So the strategy we choose analytic or non -analytic can actually be influenced by the instructions we're given.

Absolutely.

The doctor -policeman faces study by Kemler Nelson showed this perfectly.

Remind me of that one.

She created these artificial faces.

The rule was simple, long -nose -ment doctor, short -nose -ment policeman.

But she also built in an overall family resemblance for each group, hair, ears, et cetera.

And the test faces created a conflict.

A face might have a long doctor nose, but the overall look of a policeman.

And it all came down to instructions.

The group just asked to learn to recognize the faces, the non -analytic group.

They mostly went with the overall family resemblance.

They ignored the single rule.

About 60 % of the time, yeah.

But the group that was explicitly told to find the rule, the analytic group, they were much more likely to use the nose length.

It's a beautiful demonstration of our cognitive flexibility.

Okay, let's circle back to schemata for a minute, but a specific kind,

scripts.

The ones for routine events.

Right, a script is just our organized knowledge about how a sequence of events usually unfolds.

The classic example is the going to McDonald's script.

We all basically know it.

You go in, order, pay, get your food.

Eat, throw away your trash, and leave.

And we all agree on the order and on the level of detail.

We say, eat the food, not chew 32 times.

It's a cognitive shortcut.

And Bauer, Black, and Turner found that these scripts have a really powerful effect on our memory.

They do, they found two big memory errors.

First, if you tell people a story with the events all mixed up, when they recall it later, they'll put it back in the correct scripted order.

So our internal schema corrects the messy reality.

It does.

And second, if you tell a story, but leave out a central event, like you described going to a restaurant but never mentioned ordering food people, will often falsely recall the part that was missing.

Wow, so the script fills in the blanks, whether they were actually there or not.

The script is a powerful default.

It makes us prioritize and remember the central expected concepts.

This drive to find an underlying structure, it brings us to a really deep, almost philosophical idea about concepts.

Psychological essentialism.

Right, the idea that we act as if things have an underlying essence that makes them what they are, regardless of how they look on the surface.

We believe a human is a human because of their DNA, their essence.

Even though we classify people day to day based on superficial things like hair color, we believe the essence is the real defining feature.

So as we gain expertise, we move from classifying based on surface features to classifying based on those deeper essential principles.

Exactly.

And this brings it all together because the type of concept determines which model works best and what we consider to be its essence.

We have the nominal kind concepts, like bachelor or triangle.

Those have clear definitions.

The natural kind concepts, things like tiger or gold.

For these, we focus on their essential, underlying structure, their DNA or molecular makeup.

And finally, artifact concepts.

Man made things like a pencil or a TV.

And for those, what's essential is their function or purpose.

The Barton and Komatsu experiment showed this so clearly.

It was so elegant.

They gave people natural kinds, like water and artifacts, like a TV, and described transformations.

Like a TV with no picture or water that wasn't H2O.

And the results were clear as day.

For natural kinds, people cared most about the molecular change.

If water wasn't H2O, it wasn't water anymore.

But for artifacts, they cared most about the functional change.

Right.

If a TV couldn't show a picture, it wasn't a TV anymore, even if it looked the same.

It's a perfect demonstration that what we consider essential is totally dependent on the category itself.

Hashtag tag outro.

What a fantastic journey through, really, the backbone of how we think.

To recap this deep dive, we started with that rigid classical view.

With its necessary and sufficient features.

And we saw how that just crumbled when faced with real -world data on things like typicality effects and fuzzy boundaries.

That pushed the field to develop more flexible, similarity -based models, like the prototype view, with its family resemblance, and the more radical exemplar view, which stores every individual instance.

Both do a much better job with real -world variation, though they have their own issues with boundaries and just the sheer volume of information.

And finally, we moved to the really complex, explanation -based models, like scripts and the knowledge -based view, which argue that categorization isn't about similarity.

It's about our theories and knowledge that justify why things belong together.

And the most important takeaway, I think, is that our cognitive system is just profoundly flexible.

What kind of concept we build depends on the material, the instructions, and what we plan to do with the information later.

It's all about that constant trade -off.

Finding the balance between cognitive economy -saving, mental energy, and informativeness be useful for navigating the world.

And that leads to a final thought for you to consider based on what we learned about implicit non -analytic learning.

We saw that for complex things, it's often better to just absorb examples rather than trying to force a strict classical rule onto them.

So think about the last time you had to learn some new piece of technical jargon or a new skill.

Did you start by searching for a rigid, perfect definition, or did you just sort of watch and absorb a few examples in context and build a working, prototypical idea from there?

The way you approach new knowledge analytically searching for roles, or non -analytically absorbing examples, determines the kind of mental structure you build.

Which strategy do you rely on most, and is it really serving your learning goals?

Absolutely something to mull over.

Thank you for joining us on this deep dive into the very structures of human thought.

We hope you feel thoroughly informed.