Oddly Influenced

Welcome to Oddly Influenced, a podcast for people who want to apply ideas from *outside* software *to* software. Episode 50: a nugget about the preferred level of abstraction

Well, hell. In 1980’s Unix, there was a command-line program called `fortune`. You could set it up so it’d give you something catchy on your terminal when you logged in. One that stuck in my mind was quote “The first law of Frisbee: never make any prediction more specific than ‘watch this!’”.

So: I violated the law. I predicted this would be the last episode on metaphor. It’s not. It’s not because it isn’t about metaphor. Instead, it’s about a seemingly-unrelated idea in cognitive science: that, while we humans are super-fond of hierarchies or taxonomies, we are also most comfortable with a specific level in the hierarchy, what scientists call the “basic level.”

An interesting topic in isolation, with some relevance to metaphor. It will be part of my grand summation of metaphor and its relevance to software development – I hope: keep the 1st Law of Frisbee in mind.

In this episode, I’m going to lean on both our old friend George Lakoff, specifically his 1987 book /Women, Fire, and Dangerous Things: What Categories Reveal About the Mind/ and also on George Murphy’s 2002 /Big Book of Concepts/. That last sounds like a children’s book, but it’s actually closest to a somewhat dry academic literature review.

≤ music ≥

If I gave you a pile of photographs of furniture and asked you to group like with like, you might leave them untouched and say “they’re already grouped; they’re all furniture.” That would be rude. More likely, you’d group tables with tables, chairs with chairs. You *might* make three groups: tables, rocking chairs, and other chairs, but probably not: it would be weird to make distinctions among chairs but not among tables.

Tables and chairs are at what’s often called the “basic level of abstraction (or categorization).” Our brains seem to think that level special.

1. It’s the level that’s learned first. Children learn “cat” – or “kitty” – before “animal” and “mammal” and also before “Maine Coon” or “Angora.”

2, The name of the category is typically a short word. Like “cat”, so famously short that it occurs in perhaps every beginning English-language reader ever published.

3. When shown a picture and given a free choice to say what it is, people tend to answer with the basic-level name. In one experiment, people gave the basic level name 99% of the time. I don’t know what exactly they were shown a picture of, but let’s suppose it was a bulldog. So, when shown a picture of a bulldog, 1,595 said “dog,” 14 said “bulldog,” and one said “mammal” or “animal.” That experiment was an outlier, though: it’s probably more typical for about 70% of respondents to use the basic-level name, with the vast majority of the rest being the more concrete choice, like “bulldog.”

The basic level isn’t something out there in the world. Rather, it’s imposed on the world by people. For example, to a suburban boy like me, “tree” is the basic level word. That’s what I’d say if we were walking around and you asked me to name the things around me. But Tzeltal speakers in the Chiapas region of Mexico in the late 1960s were less abstract: they used words like “oak” and “maple”, one level more concrete than “tree”. They did that because trees were more important to them; they attended to them more and so could recognize important distinctions that don’t matter to me.

We can call the basic level the default choice. For example, if it’s autumn and the ground around a tree is littered with acorns, even I will know it’s an oak tree. But I’ll likely still say it’s a “tree.” To me, that’s the word I’d think most useful to a listener.

One way in which we identify basic-level objects is by part-whole relationships. I can tell a tree apart from a bush because trees have a straight up-and-down piece (the trunk) that’s really prominent. It’s notably thicker in relationship to branches than anything a bush has, it’s less obscured by leaves, and so on. Chairs are chairs because they have backs and seats, which are flattish parts at something like a 90 degree angle to each other. Even a bean-bag chair has a seat and a back; it’s irrelevant that those parts are made by someone sitting in the chair and scooching around to shape it, rather than being built that way in the first place.

Tzeltal speakers presumably key in on properties of trees that I don’t see. It could be something like leaf shape, though honestly, looking at pictures of oak leaves and maple leaves for this episode, I saw a lot that could belong to either. It could be leaf arrangement: Wikipedia tells me that oak leaves are “spirally arranged” whereas maple leaves have an “opposite leaf arrangement.” I have no idea what that means, but I expect that a tree person can tell the difference from a distance, and at a glance. Indeed, a property of the basic level is that classification is fast. Lakoff says that basic level objects are perceived as “gestalts,” wholes recognizable without first having to reduce them to their parts.

In his book /Thinking: Fast and Slow/, Daniel Kahneman divides mental processing into two… parts… or “systems”: One is fast and automatic; the other is comparatively slow and effortful. Perception of the basic level seems to be fast, suggesting it’s done by System 1, whereas other levels – which I’ll get to in a moment – are slower, suggesting System 2.

Basic level identification isn’t just perception, though. Use also matters. A handle isn’t just a long, thin thing attached to another part or to some whole: a handle has to be grippable. By us. The difference between a table and a chair isn’t just that a table has no back; it’s that the flat part of a chair is the right size for sitting on, and is set at the right height for flopping down into.

Lakoff leans heavily on how learning basic categories depends on interaction. Consider that the part-and-whole relationships of a cat and a dog aren’t wildly different: four legs, head on a flexible neck, similar shaped heads, tail, and so on. Why aren’t they lumped together as the basic category of “pet”? The answer is simple, in one sense: parents don’t teach children about pets as a category early on: they specifically point out and name “doggie” and “kitty.” But what kind of reason is that? Parents don’t use “pet” because *their* parents didn’t use “pet”? *Why* don’t any of these parents use “pet”? And what helps the child distinguish cats from dogs?

So consider a small child’s prototypical interaction with a cat. It very likely involves a nearby guardian luring the solitary cat closer, then telling the child “now, be very gentle petting her.” Cats don’t like roughhousing.

In contrast, a child on my street will almost never see an unaccompanied dog. City ordinance says dogs need to be leashed. So that’s one difference from cats: dogs come attached to people. And the words the guardian speaks are different: they’re addressed to the dog’s servant, and probably sound like “Is it OK…”, to which the servant will say something like “Oh sure, she’s very good with kids.” Such a dog will likely be very much bouncier than a cat and interact completely differently.

This understanding of how the two animals are encountered is part of their basic categories because it’s important to decide quickly how to interact with that four-legged, two-eared, thing-with-a-face over there. The category “Pet” wouldn’t be helpful. It’s such expectations that drive the closer examination of parts and part-whole relationships that leads to cats being so obviously different from dogs as a matter of gestalt.

That is, in the terminology I prefer, it’s not the visible structure of the animal that’s important, with associations to behavior tacked on. The *how* of interacting comes first, and parts are learned in support of that.

≤ short music ≥

What of the levels above and below the basic level? Classifying objects into those levels is slower – more effortful, in Kahneman’s term. That is: “Is this a mammal?” takes longer to answer than “is this a dog?”, presumably because “Mammal” is more abstract than dog. The same difficulty is present for more concrete categories. Asking “is this a Bernese Mountain Dog?” takes longer than “Is this a dog?” even among people who know what a Bernese Mountain Dog looks like.

It seems plausible that identifying a more-concrete-than-basic category takes longer because it involves attending to what Lakoff calls the “distinctive features” of the animal, which I interpret as meaning the distinctive features of its parts. For example, Wikipedia says this of the Bernese Mountain Dog:

“The Bernese mountain dog is a large, heavy dog with a distinctive tri-coloured coat, black with white chest and rust-coloured markings above eyes, sides of the mouth, front of legs, and out around the white chest. However, it is the only breed of Sennenhund dogs with a long coat.”

To put it in more mechanistic terms than either Lakoff or Murphy, there seems to be a two-step process. Recognizing that you’re confronted with a dog is immediate. But challenged to identify the breed, the brain has to play something like a game of 20 Questions with the visible features. For me, I’d classify one of those dogs based on what I noticed of the only examples I’ve ever seen, those at the Notchland Inn in Crawford New Hampshire, where Dawn and I used to stay when snowshoeing in the White Mountains. Recommended. Five stars. To this non-dog-person, Bernese Mountain dogs are: kinda bulky like a St. Bernard, with the same kind of fur, but smaller and mostly black-and-white with some brown.

Categories more abstract than the basic categories are even tougher than the more concrete ones. Children, for example, have trouble learning them. Perhaps part of the problem is that they generally have no prototypical examples to visualize.

If I ask you to picture a bird in your head, it’s likely to be an ordinary bird you can find in your yard. Perhaps a small brown songbird or, where I live, an American Robin or the Northern Cardinal (the state bird of Illinois).

Similarly, if I ask you to picture a dog or a cat, you can probably easily do it. But asking you to picture a “mammal” seems weird. I’d guess there’s a little bit of mental stutter while you think of a particular animal – a cow, say – and then visualize *that*. And I think there’s a fair chance that if I asked you to visualize a mammal this time next year, you’d pick a different one, a dog, perhaps. In contrast, if I ask you to visualize a cow now and next year, it’s likely to be the same breed each time. If you picture a female Holstein – the classic dairy cow in the US – this year, you’re unlikely to picture an Angus bull next year, though I admit that’s just a hypothesis of mine, not something I’ve seen in the literature.

Another problem with more abstract categories is that we tend to think of them only when confronted with collections of basic-level objects.

More than once, Dawn has asked for help moving a couch. She never points at a couch and asks if I can help move that “furniture.” That would be weird. She says “couch.” However, if she decides to paint a room, she’ll ask me to help move the furniture out of it: the collection of chairs, couches, and the one, uh, gentrified lobster trap Dawn brought from New Hampshire to Illinois that sits next to my favorite chair and holds the books I use in episodes like this.

≤ short music ≥

It’s easy when reading books about categorization to think everything’s quite tidy, but – as constant listeners know – I’m disinclined to think of anything in the brain as being *tidy* in the way that programmers or non-biology scientists think of tidiness or elegance or simple or any of those magic words.

Consider my brain on dogs. I’m not a dog person, just like I’m not a tree person. If you gave me a free choice classification test, you’d probably conclude quickly that “dog” and “tree” are basic categories for me. However, there are three breeds of dog that I recognize instantly in a way that feels quite different: as immediate as recognizing them as dogs. They’re somehow both below a basic category and a basic category themselves.

Those breeds are:

1. The Springer Spaniel. Dawn bred and showed Springers as a girl, so they’re important to me. Notice, though, that there’s an intermediate level of abstraction between “dog” and “Springer Spaniel,” that of “spaniel.” Not only can I not immediately identify some random dog as a spaniel, I can’t even tell you anything about the difference between spaniels and, say, terriers. That’s a level of abstraction that has never mattered to me, so it might as well not exist.

2. The Corgi. Corgis are pretty recognizable because they have embarrassingly short legs for their body. However, I suspect it matters more that our street is positively infested with Corgis. And even more than that, there’s a Corgi that lives across the street and two houses down. Pretty much every evening while Dawn and I are sitting on our front stoop, Penny is sitting in our neighbors’ biggish front window and looking at us. She’s *salient* to my life. Dawn and I point her out to each other fairly often, so my brain recognizes her breed, and fast.

3. The German Shepherd. I’m not visual, so I have few visual memories. One, though, is a childhood memory of a German Shepherd attached by its jaw to my leg. 36 stitches later and some time on crutches, you best believe I recognize German Shepherds without thinking about it.

My point is that it’s tidy to think of taxonomies as having abstract, unvisualizable levels above the basic level, a basic level that features fast, effortless recognition of a gestalt, and – finally – more concrete levels that require effortful discrimination among alternatives. In this model, all the levels are distinct from each other, are different *kinds* of levels. But that’s imposing our preferences on a brain that doesn’t care about such proprieties.

Thank you for listening. Next – here I go making predictions again – an interesting nugget about the construction of memory.