Oddly Influenced

Welcome to Oddly Influenced, a podcast for people who want to apply ideas from *outside* software *to* software. Episode 46: How do metaphors work?

“I’m back.” No excuses, sirs and ma’ams, for the long absence.

With this episode, I’m starting a series roughly organized around ways of discovering where your thinking has gone astray, with an undercurrent of how techniques of literary criticism might be applied to software documents (including code).

This episode is about metaphors, how they influence problem solving, and how you might work with that to better notice when your solution isn’t right. I’ll be leaning on four essays from the second edition of /Metaphor and Thought/, a 1993 collection edited by Andrew Ortony. Those essays are:

* “The conduit metaphor: A case of frame conflict in our language about language,” by Michael J. Reddy,

* “Generative metaphor: A perspective on problem-setting in social policy,” by Donald A Schön.

* “Metaphor,” by John Searle, and

* “The contemporary theory of metaphor,” by George Lakoff.

I have too much content for one episode, so this episode will discuss theories of metaphor, saving what those theories might have to do with software practice for the next episode. I’ll end with my own special not-entirely-crackpot theory, which is closest to Lakoff’s but informed by some of the previous episodes on ecological and embodied cognition.

If you like Lakoff’s theory (in the original or in my variant), a book for you to read is /Metaphors We Live By/, by George Lakoff and Mark Johnson. It’s an easyish read and not too long. I have the 1980 edition. There’s since been a second edition, but I haven’t read it.

≤ music ≥

Here’s a made-up story. Dawn’s favorite part of the immune system is the helper T-cell. They’re the cells that pass on news of an infection to other cells that can deal with it. The process is ridiculously complicated. Suppose one day Dawn was explaining it to me. When she finished, I said, “That went completely over my head.”

That’s a metaphor: two, actually. First, I compared the idea she was expressing to an object moving through space. The fact that it went *over* my head means the idea didn’t get *into* my head. In fact, it went *completely* over my head, implying that it could have left some partial understanding behind, but it didn’t.

That seems like a crazy way to say “I didn’t understand anything you just said.” In his essay for the collection, John Searle explains how he thinks metaphors work and why we use them. He gives a fairly mechanistic process.

First, normal conversations have rules. One of the rules is that what you say should be relevant, isn’t a non-sequitur. So Dawn’s brain, hearing “That went completely over my head,” has a “huh?” moment. “Why is Brian babbling about things flying over his head?” Because I’m a norm-following conversational partner – that’s what first attracted Dawn to me – her brain knows that I mean to convey to her something that’s true and relevant by using something untrue and irrelevant. Essentially, I’ve set her a puzzle to solve. She solves the problem by thinking about properties of the fictional situation, and wondering how similar properties might apply to the actual situation.

So, “over my head” means “not in my head,” which implies “not in my brain,” which means the message wasn’t received, which means understanding wasn’t achieved.

Searle devotes only one paragraph of a 28 page essay to explaining why I would want to make Dawn go through all this.

“The expressive power that we feel is part of good metaphors is largely a matter of two features. The hearer has to figure out what the speaker means – he has to contribute more to the communication than just passive uptake – and he has to do that by going through another and related semantic content from the one which is communicated. And that, I take it, is what Dr. Johnson meant when he said metaphor gives us two ideas for one.”

Boom, end of essay

In other words, I’m giving Dawn work to (1) make her pay attention and (2) to remind her that objects can move through the air roughly horizontally, and sometimes pass above a fixed point of reference, such as a head.

The first seems rude, and the second pointless, so I don’t buy it. But let’s leave the “why say it?” question open for now.

In his essay, Lakoff has roughly the same explanation for how we understand a metaphorical statement but a quite different explanation for what happens next.

A main difference in Lakoff’s story about understanding is that he deals in metaphorical systems rather than just individual sentences. For example, consider the following sentences:

Look how far we’ve come since that first date.
Our marriage is at a crossroads.
This relationship isn’t going anywhere.
We may have to go our separate ways.

The metaphorical system is that love is a journey. Lakoff expands on it like this:

“The lovers are travelers on a journey together, with their common life goals seen as destinations to be reached. The relationship is their vehicle, and it allows them to pursue those common goals together. The relationship is seen as fulfilling its purpose as long as it allows them to make progress toward their common goals. The journey isn’t easy. There are impediments, and there are places (crossroads) where a decision has to be made about which direction to go in and whether to keep traveling together.”

There are two of what Lakoff calls “domains” here. The *source domain* is a physical journey, and the *target domain* is a love affair. You could draw both of them as graphs. For the source domain, the nodes would be labeled with nouns like “vehicle” or “destination” or “journey” itself. Nodes would have properties. For example, the “journey” node has a property “not easy”. Arcs between nodes signify relationships. For example, the vehicle moves toward the destination.

You could draw a similar graph for the target domain, using nouns like “lovers,” “relationships,” and so on.

Finally the graphs can be combined by drawing lines between some (not necessarily all) of the elements of the two graphs. So there’s a line from “lovers” in one domain to “passengers” in the other. Lakoff is using the mathematical metaphor of a “mapping.” (He doesn’t actually draw such graphs in this material, but I’ve seen them elsewhere.)

Searle gives the impression that the meaning of a sentence is calculated; whereas Lakoff is more in the style of a lookup. So when you hear “our marriage is at a crossroads,” you look up “crossroads” in the source domain, follow a pointer over to the corresponding element in the target domain,” and read off its meaning.

That’s not to say you might not do a fresh calculation. For example, Dawn and I were engaged to be married four months after we first met. If I say “we had a whirlwind romance” and you’d never heard that idiom before, you could use the “love is a journey” metaphorical system to interpret it.

Conclusion: we were not in control of our vehicle.
Conclusion: we were being carried along much faster than expected.
Conclusion: we were lucky to have made it to our destination safely.
Conclusion: if the relationship had ended, it would have ended spectacularly.

And so on.

It’s reasonably likely the idea of a “whirlwind romance” would stick in your head – in effect, you’d have cached the results of your calculation in the metaphorical system itself.

The key difference between Searle and Lakoff, it seems to me, is when you move beyond understanding to problem-solving. If I understand Searle correctly, here’s what would happen with my “That went completely over my head” statement:

First, as before, Dawn would solve the verbal puzzle I’d posed to her. It sounded like I was saying something flew over my head: a truth-valued statement that was literally false. But what I intended was a true statement that I hadn’t understood her. Now she’s faced with a decision: what to do about it? The reasoning she uses to find a solution has nothing to do with the metaphorical statement, only with my intended meaning. Problem-solving happens in the target domain.

Lakoff thinks it’s far more likely that the reasoning happens in the source domain, that you solve problems with metaphors. In his essay, Schön gives an example that I quite like.

“Some years ago, a group of product-development researchers was considering how to improve the performance of a new paintbrush made with synthetic bristles. Compared to the old natural-bristle brush, the new one delivered paint to a surface in a discontinuous, ‘gloppy’ way. The researchers had tried a number of different improvements. They had noticed, for example, that natural bristles had split ends, whereas the synthetic bristles did not, and they tried (without significant improvement resulting) to split the ends of the synthetic bristles. They experimented with bristles of different diameters. Nothing seemed to help.

“Then someone observed, ‘You know, a paintbrush is a kind of pump!’ He pointed out that when a paintbrush is pressed against a surface, paint is forced through the *spaces between the bristles* onto the surface. The paint is made to flow through the ‘channels’ formed by the bristles when [those] channels are deformed by the bending of the brush. He noted that painters will sometimes “vibrate” a brush when applying it to a surface, so as to facilitate the flow of paint.

“The researchers tried out the natural and synthetic bristle brushes, thinking of them as pumps. They noticed that the natural brush formed a *gradual curve* when it was pressed against a surface whereas the synthetic brush formed a shape more nearly an angle. They speculated that this difference might account for the ‘gloppy’ performance of the [second] brush. How then might they make the bending shape of the synthetic brush into a gentle curve?

“This line of thought led them to a variety of inventions. Perhaps fibers could be varied so as to create greater density in that zone. Perhaps fibers could be bonded together in that zone. Some of these inventions were reduced to practice and did, indeed, produce a smoother flow of paint.

Paintbrush-as-pump is an example of what I mean by a generative metaphor.”

This is by no means irrefutable evidence that reasoning was done in the source domain. Following Searle, we could say the metaphor was interpreted as “paint flows between the bristles, not through them, so pay attention to that space” and thereafter everyone forgot about pumps. But I doubt it. First of all, I bet they didn’t just drop the metaphor but kept coming back to it. And I bet they came back to it to ask “what would make a pump deliver water in an annoyingly spurty way?” And made observations like “if a pump’s channel had a sharp angle in it like the ones we sort of see in brushes, that might lead to discontinuous flow.”

Still, we need more evidence before believing Lakoff. Reddy’s essay on what he calls the “conduit metaphor” provides some and is interesting in its own right. It was actually the founding document for the what’s typically now called the “conceptual theory of metaphor” (rather than Lakoff’s “contemporary theory”). (That, for example, is the title of the wikipedia page.) Reddy was also the pioneer of a part of its explanatory style, which is to overwhelm you with examples until you give in and say, “OK, OK – metaphors are *everywhere*. They must be doing *something* important.”)

His essay has an appendix with 141 different types of metaphors about communication. Here are some examples I just made up:

“Try as I might, I just can’t get the idea ACROSS to her.”
“I’m TAKING the idea of the conduit metaphor FROM Reddy’s essay.”
“Jerry Weinberg said consultants should GIVE AWAY their best ideas for free.”
“It’s hard to PUT the idea INTO words.”
“The ideas COME too fast and furious to follow.”
“His aside was PREGNANT with meaning.”
“That claim is WITHOUT meaning.”
“You OVERLOOKED the whole point of the essay.”
“You need to get your ideas OUT THERE, where they might do some good.”
“Get her started, and the ideas will just POUR OUT OF her.”
“Let me SET my thoughts DOWN ON paper.”
“As I drove Dawn to New Orleans for her board certification exam, she was CRAMMING in the passenger seat the whole time.”

The metaphorical system underlying these examples goes like this:

* Someone – a speaker or writer – has an idea in her head.
* She encodes it into a message, conceived of as a container for ideas.
* That message is then either sent to a specific person – like shooting it down a conduit – or it’s broadcast for any number of people to receive.
* The message is then decoded so that the idea is now in the receiver’s head. This decoding is the inverse of the encoding.

Reddy’s evidence that people are really using this understanding – not doing a Searle-like translation to literal truth statements – is that our reasoning about communication is often *bad*, and it’s bad in ways consistent with use of this metaphor. He describes two badnesses.

For the first, I’ll work from something the poet and programmer Richard P. Gabriel once said to me: “A poem is a program that executes in the brain of the reader.” To me, that highlights – and forgive me here, pure functional programmers – the importance of mutable state. The text of the poem interacts with the mental residue of a reader’s lifetime of experience. That interaction produces a meaning or interpretation that will differ in at least some small way from every other reader’s.

Meaning is not put *into* the poem, or into any text; it is constructed by the receiver’s brain in response to clues delivered as words. We communicate successfully because we share experiences in common – and thus have a sort of shared state with our fellow humans – and because we’re trained from an early age in how to interpret words so that our meanings are close enough to other people’s for our life’s purposes.

Because the conduit metaphor ignores the receiver’s context, Reddy claims we tend to also. We do a poorer job than we should in anticipating our reader’s context, and then we blame them when they don’t interpret our words the way we think they should. The message is *right there*, in the words – how can they miss it?

That leads to the second badness, which is the effect of a metaphor embedded within the conduit metaphor, that of putting ideas into containers that someone else takes them out of. Reddy points out that our practical experience with containers is that the taking-out is easier than the putting-in. Consider a Christmas present: you cut up a sheet of wrapping paper, discover it’s too small, cut out another, try to get the stupid wrapping paper to fold neatly, put on a ribbon, put on a bow, write a card with something clever on it. And then the recipient undoes all that work in a matter of moments.

That accounts for the frustration when the listener doesn’t get the message. Extrapolating from experience with physical containers makes it seem like that should be easy. Failure to understand is frequently attributed to the reader’s perversity, with statements like “You’re reading way too much INTO that.” (That is, the reader is replacing the original content of the message with a different content.)

Meanwhile the listener is inclined to think the speaker packaged the message wrongly. Like my mother – rest her soul – who consistently and cheerfully went way overboard with the scotch tape, making her presents “cutely” hard to unwrap.

On average, it seems that the speaker gets most of the blame, since there are more metaphors that talk about shoddy construction of the message than shoddy reading. Some examples:

“He tries to say way too much in every sentence.”
“You just dumped every last idea you had into that podcast episode, didn’t you?”
“Simple ideas hiding behind a thicket of jargon.”

And our downplaying of understanding as a potentially difficult task has practical consequences. We spend more time teaching people how to write than how to read well. Careful, critical reading is a specialized skill most people don’t learn and do not practice. My hope for this series is that I’ll teach you a bit about different ways to read.

Does this clinch the case that problem-solving is metaphorical? Naw. I don’t see how you can definitively defeat the idea that problem-solving happens only in the target domain by any means short of watching which neurons where do what when stimulated by a metaphor, and I haven’t made the effort to hunt down if anyone’s tried to do that. Instead, I’m willing – as is this podcast’s habit – to credit a theory and see what it suggests for software.

≤ music ≥

But first.

I’ve been in a mixed marriage for going on 36 years. I’m a computer person. Dawn is a medical person. My profession is obsessed with minimalist, highly structured or even formal solutions that can be readily comprehended or summarized. In contrast, Dawn has impressed upon me that the body doesn’t give a tinker’s damn about any of those properties. Your body is a gross kludge. And the same goes for your brain.

So I’ve been acculturated to be highly suspicious of anything like an elegant or logical solution when it comes to the body or brain. Therefore Searle’s “language is about boolean-valued propositions” strikes me as wildly implausible, even as what he calls “a rational reconstruction of the inference patterns that underlie our ability to understand such metaphors.” I bet he’s reasoning backward from his conclusions. Or perhaps his desire that reality be a certain way.

Lakoff is more to my taste, but that image of neat little source and target domains with mappings between their elements seems to me very likely to be missing important things by abstracting away the inherent messiness of our brain.

So let me offer an interpretation of metaphor that is broadly compatible with Lakoff but that I think has some additional virtues. I should emphasize that I’m in no way qualified to be inventing a theory of the neurobiology of language, but I don’t think I’m being *too* irresponsible to the ideas of better qualified people.

I’m inspired by the episodes that are numerically just before this one, but are also more than a year old. So I’ll do some recapitulating.

The brain does its work by building networks of neurons in response to repeated stimuli. So a rabbit, for instance, has a set of neurons in its olfactory bulb (a part of the brain) that have been exposed to the smell of a carrot, say, multiple times in the context of food. Learning the smell means physical changes to those neurons that make them tend to fire together in a self-sustaining pattern. A network has been created, tailored to the smell of carrots.

This carrot-recognizing network takes signals from perceptual neurons. If those signals correspond to the smell of a carrot, the network becomes activated. Repeated activations – caused by the way rabbits go sniff-sniff-sniff all the time – increase the activation’s strength until a threshold is reached, which causes the network to send a signal to the cortex, where other networks arrange for the rabbit to do something about the smell.

It’s important to realize that your brain has *lots* of room for networks. You have about 100 billion neurons. They’re densely interconnected, so there are about 100 trillion neural connections. There’s no way to imagine such a huge number, only essentially meaningless comparisons like “250 times as many connections as there are stars as in our galaxy” or “if you counted one neuron a second, it’d take 3,171 years to count them all.” I can’t imagine 3000 years any better than I can 100 billion neurons.

Also, when thinking of networks, it’s a mistake to think of them as discrete and isolated – like what you might imagine for the little “love is a journey” graph I described earlier. Better to think of an individual neuron as doing work for multiple networks and of networks overlapping each other so much that the idea of “boundaries” is a quaint fantasy. Activate the “love is a journey” network, and who knows what other networks will be partially activated?

The philosopher and cognitive scientist Andy Clark thinks that concepts are networks that are anchored by words. That is, there’s a network for the word “gulf.” That network doesn’t correspond to a dictionary definition like “a large inlet from an ocean into a landmass,” although someone with appropriate experience could build such a statement from the networks in their head. Instead, the “gulf” network is linked to all kinds of random things, like the memory of the body of water you lived on as a kid, the Gulf Oil company that had a lot of gas stations around when you were growing up, some representation of the difference between a gulf and a bay, and so on. And now, because some weird dude with clout claims to have renamed the Gulf of Mexico to the Gulf of America, *that* nonsense will probably be associated the word “gulf” in my brain until it finally gratefully gives out.

When one hears a word – or thinks it – networks associated with the word light up to varying degrees. Thus it is that I can rarely read the word “dawn” – meaning the beginning of the day – without thinking of Dawn the person. Although “think” isn’t necessarily the right word. It feels like a little flash in her direction, which I like to think of as the physical effect of an important association lighting up.

Although Clark talks of words anchoring content, I expect phrases work just as well. That is, the meaning of “whirlwind romance” is stored independently – has its own network. It’s its own thing. It’s associated to the meanings of “whirlwind” and “romance,” sure, but it’s not *dependent* on them.

≤ short music ≥

If you read up on metaphors, you’ll likely see a claim that there’s such a thing as a “dead metaphor.” A dead metaphor, per wikipedia, is “a figure of speech which has lost the original imagery of its meaning by extensive, repetitive, and popular usage, or because it refers to an obsolete technology or forgotten custom. Because dead metaphors have a conventional meaning that differs from the original, they can be understood without knowing their earlier connotation.”

The implication is that dead metaphors are, using my earlier metaphor, just a quick cache lookup of the literal meaning, without any calculation happening at all. In that they differ from the pump metaphor or Shakespeare’s:

But, soft! what light through yonder window breaks?
It is the east, and Juliet is the sun.
Arise, fair sun, and kill the envious moon,
etc. etc.

… which provokes thought, interpretation, or decision-making, and – moreover – threads through a whole scene.

But can a metaphor be really dead? Let’s use as an example the phrase “balls to the wall.” The literal meaning is “all-out effort.” Its origin, according to two sources I’ve seen, is in aviation. Here’s what Wikipedia says:

“First attested in the 1960s in the context of aviation, in reference to ball-shaped grips on an aircraft’s engine control (typically throttle, prop pitch, and fuel mixture). Pushing these ‘balls to the wall’ would put the aircraft at maximum thrust.”

(The quote doesn’t say, but the “wall” here is the plane’s control panel in front of the pilot. A correspondent told me that, in the full forward position, the throttle would actually touch the panel: balls to the wall.)

Now, it’s easy to imagine aviators inventing that catchy, semi-rhyming term from their own experience flying, then using it metaphorically in other situations calling for maximum effort, and having the phrase spread until now it’s used by people who have no clue about its origin.

But really, who cares? What matters is whether and how it affects decision-making in people who use or hear it. I claim this dead metaphor does affect decision-making, at least for me. That makes it not dead, I think, at least in any way that matches the metaphorical meaning of the word “dead.”

I have no idea where I first learned the phrase “balls to the wall,” but I bet it had nothing to do with aviation. (I didn’t know about that origin until this last month.) I expect that I learned it in a context where someone was praising someone else for stepping up, going beyond expectations, making unusual effort. Later repetitions in similar contexts strengthened its network so that the phrase now comes easily to mind and is easily interpreted.

Now, the meaning was never computed from the meanings of the word “balls” or the word “wall,” but when the network for “balls to the wall” is activated, networks associated with those words are also activated, I believe. Now, I’m an American male, which means that – outside of sports – the default meaning of “balls” is testicles. And that means “balls to the wall” is infected with the connotations of that organ, which are mainly around courage, as in “That was a ballsy move,” or “He didn’t have the balls to jump off the highest diving board.” A secondary connotation is aggression, specifically the stereotypically masculine energy and aggression that Mark Zuckerberg recently said has gone missing from tech companies like his.

The way I think of it is that activating the network for “balls to the wall” activates networks related to courage, aggression, and masculinity, making those concepts more available to decision-making.

Aggression is also reinforced by the connotations of motion toward a “wall.” There are thousands of TV episodes where the police push someone against a wall and frisk them in a not-gentle way. I doubt there’s more than a handful of people who don’t make that connection when they hear the lyrics to the song “Up Against the Wall Redneck Mother.”

So when someone says, “if we’re going to make this deadline, we have to go balls to the wall,” they don’t mean “we have to be thoughtful in dealing with obstacles because going faster entails more risk,” and no one thinks they do. In fact, they’ll find it harder – if Lakoff is right – to even think about risk other than as something to embrace, like Real Men do.

Well, except for me. And that’s because of a story Dawn told me, which I’ll now tell you. I have to warn you that I almost invariably remember such stories better than she does, so it’s possible my memory has embellished things.

Once upon a time, Dawn had a case involving a prize breeding bull. This bull was so genetically gifted that six different people pooled their money to buy him. One day, this bull took a dislike to another bull on the other side of what I remember Dawn saying was a wall, but might have been a fence. Our bull attempted to get over the barrier to fight the other bull, but only made it half way. As a result, he unfixably ruined his testicles. He was no longer a breeding bull, but a beef steer worth maybe two dollars a kilo, so the syndicate was out a lot of money.

To my brain, that bull became the definition of “ballsy.” He sure went balls to the wall. And that’s tinged my understanding of masculine energy ever since. Which has been reinforced by seeing software projects where people went balls to the wall and only made things worse. To me, “balls to the wall” sounds like the advice “When in danger, when in doubt, run in circles, scream and shout.” It’s the sort of thing that causes you to spend tens of billions of dollars on virtual reality it turns out no one wants, or fire nuclear safety workers, only then realizing, actually, keeping nuclear weapons safe *is* a legitimate function of government, and then have trouble un-firing them because – oopsie – you don’t have their contact info any more. To name just two ballsy decisions made by people who fancy they have masculine energy.

I can no longer hear of “masculine energy” or think of ballsy initiatives without remembering that big, dumb bull who wouldn’t stop to think because, to him, aggression was self-justifying.

Because Dawn’s story was so memorable – except to Dawn, I guess – it rather effectively altered my network in a consequential way. In contrast, learning the origin of the metaphor “balls to the wall” made a tiny tweak to its network that I doubt will ever affect anything.

I conclude that the metaphor of a “dead” metaphor is not useful. It in fact leads us astray because it downplays the role of ordinary metaphors in problem-solving.

≤ short music ≥

That’s it for theory. Next episode will be a shorter one with some suggestions for what you can *do* with this information. Thank you for listening.