What the F

by Benjamin K. Bergen

  People make fewer speech errors when the result would be taboo.

  But people are constantly not making errors. We know that people make fewer errors producing taboo words, but that doesn’t mean that every time they successfully avoid an error they’re making a correction. Some proportion of the time, they’re probably just getting it right from the outset. We need a way to diagnose whether people are activating an internal plan to say something that they eventually don’t produce. This is nearly impossible to do.

  Except with taboo words. That’s because even thinking about taboo words has special effects on the human body. When people say taboo words, their pores open up within seconds, and they sweat. And this is measurable.18 Sweat conducts electricity, and the more sweat there is on your skin, the more conductive the surface of the skin will be. So, you can pass a very low level of electrical current across people’s skin, say, on a finger, and measure how conductive the skin is. When people start sweating, the conductance increases. This is the basic logic behind what’s known as the galvanic skin response (GSR). You might be familiar with this tool because it’s one of the components of a traditional lie-detector test—skin conductance also changes as a function of anxiety, which may be driven by lying in some people.19 Critically, although many words make people sweat, including emotional words like murder and hate, the most profuse and most reliable sweating comes from hearing taboo words.20

  Suppose you were able to measure skin conductance as people performed the word-pair reading task. If merely planning a taboo word—even one that ultimately gets internally corrected and is therefore never pronounced—makes people sweat, then this should show up as an increase in skin conductance.

  So here’s the idea. You have people produce word pairs. Some, like tool kits, can induce profane errors, which, most of the time, people successfully avoid making. But looking just at the times when people didn’t make an error, you can split these correct pairs into two groups. The first includes those where the participants’ skin conductance spiked—suggesting that they had internally activated a taboo word, even though they didn’t make an error. And the other group includes instances where there was no error and also no spike—suggesting that no profane word was ever even considered. And you look to see if the sweaty trials also take longer. Together, an increase in skin conductance and a longer time to produce the word pair would provide compelling evidence—albeit still circumstantial—that people were internally planning, but ultimately taking the time to correct, taboo words.

  And that’s what was found. The chart on the next page shows how long it took people to successfully avoid taboo errors when they had large and small GSRs.21 As you can see, when they were sweating (the high-GSR group), they also waited significantly longer to start talking (the left bar is taller).

  Here’s another, lower-tech way to detect whether people are doing active editing before they speak. It has to do with the different types of errors people make. We’ve already talked about anticipatory errors (hit shed becomes shit shed) and exchange errors (hit shed becomes shit head). But people also make something known as a perseveratory error, which takes a sound pronounced earlier and repeats it later (hit shed becomes hit head). The taboo-error word pairs we’ve been talking about, like hit shed and heap chore, typically have the potential to produce an obscene word on either the first word or the second word but not both. As a result, for any given pair, an exchange error will always produce an obscene word (hit shed becomes shit head), and so will either an anticipatory error or a perseveratory error, but not both. In the case of hit shed, an anticipatory error produces the obscene shit shed, but a perseveratory error gives you hit head, which isn’t so bad. With pairs in which the taboo word would be in second position, like heap chore, only the perseveratory error (heap whore) and not the anticipatory one (cheap chore) would be obscene. So if people are editing, then when the taboo words would be first, people should avoid the taboo word by making more perseveratory errors and fewer anticipatory ones. And when the taboo word would be second, this should reverse: people should make more anticipatory than perseveratory errors. That’s what you see in the next chart.

  People who successfully avoid making taboo errors take longer to speak when they’re sweating more (high GSR).

  What’s more, you might notice that the numbers are far greater when the second word of the pair is taboo than when the first is—the light bar on the right towers over the others. Why would this be? Why would people make more anticipatory errors when the second word would be taboo? This too seems to implicate internal editing. If editing takes time, then presumably you’d be more likely to catch and stop an error when it’s on the second word of a pair than when it’s on the first. And if the errors you’re most vigilant about are taboo ones, then it follows that you’d be most likely to catch and avoid those errors when they’re planned for the second word.

  When people do make errors, they avoid generating a taboo word.

  This work, all conducted in the late 1970s and early 1980s, was influential at the time because taboo language provided a privileged way to detect internal editing processes. More recently, researchers have been interested in tracking down the brain basis for internal quality control of language. And they’ve turned to the same foundational paradigm, with some major technological additions.

  The first new twist has people perform basically the same word-pair reading task while tethered to an electroencephalogram (EEG) machine that measures their brain waves.22 In a nutshell, here’s how EEG works. Electrodes (lots of them—as many as 256 but more often 32 or 64) are applied harmlessly to the scalp. The electrodes measure fluctuations in the electrical field, and the specific electrodes used in EEG experiments are sensitive enough to measure microvolts—one-millionth the voltage of your AA battery. A lot of things affect the electrical field measured by electrodes placed on the scalp, including passing airplanes, elevators, and even the muscles firing when a participant blinks. But it turns out that highly sensitive electrodes can detect something far more important to cognitive scientists: the activity of neurons. When a neuron fires, at the chemical level a bunch of ions flow into or out of it. And those ions carry electrical charge (those pluses and minuses attached to Ca2+ or Cl–). So when a nerve cell fires, the flow of ions affects the electrical field around it. And when thousands or millions of neurons oriented similarly and located close to each other fire at once, the electrical field change is strong enough for those sensitive scalp electrodes to measure.

  Over many decades of research, neuroscientists have observed people performing hundreds of tasks while measuring their brain waves using EEG. And they’ve observed that certain types of behavior produce predictable changes in the measured electrical field. For instance, about four hundred milliseconds after you see a word, the electrical field centered over the top of your scalp deflects negatively. This is believed to index the process of interpreting the meaning of the word and integrating it into your ongoing understanding of the language you’re reading or hearing.23 Other components of the electrical signal relate to other specific behaviors and cognitive processes.

  And so, when you wire people up to an EEG and have them read error-inducing word pairs, their brains produce different electrical signals, depending on what type of error they’re avoiding.24 A temptingly taboo pair like bunt call induces a stronger negative-going inflection over the center of the scalp about six hundred milliseconds after the prompt to speak, as compared with what happens when the same brain sees a neutral pair, like bunt hall. And this is true even when the person doesn’t actually commit a verbal error. This tells us that the brain is doing different things when successfully avoiding a taboo error versus a neutral one. It doesn’t reveal exactly what those different processes are—we only know for sure that neurons are firing differently—but it does tell us when that difference occurs. At six hundred milliseconds after the prompt to speak, people’s brain activity diverges, depending on the type of error that wou
ld be produced. This suggests that the brain is doing something different when you’re planning speech and not making taboo errors than when you’re planning speech and not making mundane errors.

  But we want to know not only whether something different is happening in people’s brains and when but also what is happening. And because the human brain exhibits localization of function—as we saw in the last chapter, circuits located in different places execute different computations—knowing the location of the brain differences revealed by EEG could help us figure out what they mean. Unfortunately, it’s notoriously challenging to extract locational information from EEG; changes to the electrical field measured at a particular electrode aren’t necessarily due to the activity of neurons located directly below that electrode in the nearest piece of tissue. (The issue is complex, but it turns out that the direction the neurons are pointing matters, among other complicating factors.)f

  The real issue is that this is a type of “inverse problem.” Here’s the basic idea. Even if you know what the output of a complex system is, tracking down its causes turns out to be impossible. This is because although the output is determined by the system, the system is complex enough that many different system behaviors could produce the same output. For more, see Baillet, S. (2014).

  But other techniques can tell us something about location. One is functional magnetic resonance imaging (fMRI), which measures fluctuations in the magnetic field from outside the body. When neurons fire, they use energy, and the more they fire, the more oxygenated blood flows to them, providing more energy (in the form of ATP, which you might remember from high school biology). Rushes of oxygenated blood can be measured by their magnetic signature, and so this can serve as a somewhat delayed and messy proxy for where neurons are firing in the brain. When you get more of this blood-flow signal to a particular region in the brain for one task than another, chances are the neurons in that region are doing more during the first task than during the second.

  Applied to the same word-pair reading task we’ve been tracking, fMRI starts to fill in the picture of what the brain is up to while it’s editing planned words. When you compare the brain’s blood-flow signal during the taboo-eliciting pairs (tool kits) and neutral pairs (tin cable), you find that they’re significantly different in one place, shown on the next page.25

  The little blob in the lower part of the frontal lobe of the right hemisphere is in a region called the right inferior frontal gyrus, which is implicated in inhibitory control—your ability to stop or prevent yourself from doing something. Suppose you’re waiting at a traffic light, for instance. It turns green, so you prepare to step on the gas, but then just as quickly, it turns red (perhaps because an emergency vehicle or train has overridden the usual light sequence). You need to quickly interrupt your plan to move forward. This appears to be the specialty of the right inferior frontal gyrus. It sends a hold-the-presses signal to stop action before it starts.26

  Is it just me, or does that sound a lot like what an internal editor would be doing when confronting a taboo word that’s about to come out of your mouth?

  # $ % !

  The right inferior frontal gyrus (a brain region involved in inhibitory control) experiences increased blood flow when people avoid taboo errors as compared with nontaboo errors.

  The evidence from taboo speech errors and what happens when you avoid them implicates an internal process of self-monitoring. You are constantly censoring your words even before you articulate them in order to avoid slipups like the pope’s. And it looks like, as far as the brain is concerned, suppressing an error—in particular an erroneous taboo word—is a lot like suppressing an action in response to an external stop signal.

  But this is only the beginning. There are other ways to tell that people call in an all-systems-halt signal when they feel they’re about to inadvertently say something obscene. We know this because psychologists have been trying to trick people into profane slipups in a variety of ways for decades.

  One way is a well-known phenomenon called the Stroop effect. Basically, if you have people look at words and say what color font they’re printed in, they do pretty well. Show them a word written in blue, and they can say it’s blue. Unless, that is, the word printed in blue ink happens to be the word red. Then it gets a lot harder—people are slower and make more mistakes. That’s the normal Stroop effect, and it’s interesting in its own right because it reveals that you can’t help but process the meanings of the words that you read, even when willing yourself to pay attention only to the color of the ink. You process meanings automatically and you’re tempted to produce them. To avoid errors, you slow down.

  Strangely, taboo words induce a Stroop effect as well: they interfere with people’s ability to say what color a word is printed in. It’s hard to demonstrate this on the printed page when you only have black ink (what century is it again?), but here’s a quick-and-dirty approximation. We’ll replace color with typography. Your job is to go through the list of words below, in order, and say whether each one is printed in italics, bold, or underline. Do it as quickly and accurately as possible. OK, go.

  rib-eye

  stethoscope

  mountain

  pitchfork

  italics

  donut

  library

  underline

  fire-engine

  barley

  philanthropy

  bold

  stegosaurus

  clandestine

  cunt

  fortuitous

  bicycle

  fuck

  momentum

  If all worked according to plan, you should have noticed that this task was harder for some words than others. The words that denote a particular font style, like the words italics and bold, should have been tough when they didn’t match how they were printed. They should have taken longer, and you might even have made mistakes on them. That’s the normal Stroop effect. But the taboo words should have taken longer as well: this is the taboo Stroop effect. You can see data from the first taboo Stroop experiment conducted on the next page, as described in a 1995 paper.27 As you can see, the normal Stroop task causes a delay of about 150 milliseconds—compare the middle bar with the leftmost “incongruent” bar. The taboo Stroop (the middle bar versus the rightmost bar) is nearly as large.

  The normal Stroop effect causes people to name colors about 150 milliseconds slower, as shown by the difference between the control and incongruent conditions. The taboo Stroop is of similar magnitude, as demonstrated by the difference between the control and taboo conditions.

  What causes the taboo Stroop? Certainly part of the story has to be the same as for the standard Stroop: it’s hard to ignore the meaning of words that you focus on. Otherwise, you could selectively attend to the style of printing, and there would be no Stroop effect for us to talk about in the first place. Taboo words of course are particularly hard to ignore. But why do taboo words cause a delay in speaking, as compared with control words? Suppose, as we conjectured earlier, that when you perceive that you might mistakenly produce a taboo word, your internal monitor hits the brakes. This would lead to longer reaction times when the information that you’re supposed to ignore (but might erroneously produce) is taboo. So the taboo Stroop effect could be explained once again by internal self-monitoring.

  To be fair, there are other possible explanations. Perhaps the most convincing one sidesteps the issue of production and inhibition entirely. We know that seeing a taboo word evokes an emotional response. As I mentioned earlier, when you measure skin conductance as people simply listen to words, even when they don’t need to speak at all, GSR is larger for taboo words than for neutral ones.28 So an alternative explanation of the taboo Stroop effect is simply that this emotional response sucks up mental resources that you’d otherwise need to name the color of the word.29 In effect, the emotional jolt you get from profanity overwhelms you to the point where other tasks you’re trying to perform concurrently get put on the ba
ck burner and therefore take longer.

  There’s some corroborating evidence for this view. Experiencing strong emotions leads people to instantaneously encode a memory of what they’re experiencing when that emotion hits, generating a so-called flashbulb memory, like the birth of a child, seeing the space shuttle explode, and so on. So if the taboo Stroop stems from a strong emotional response to seeing a taboo word in the middle of a psychology experiment, then your brain should encode an image of the event—in this case the word—that’s stronger than memories typically encoded for less emotional experiences, like neutral words in the same experiment.

  In fact, when you spring a pop quiz on people who have participated in one of these taboo Stroop experiments, they remember the taboo words they saw far better than they do neutral ones. Not only do they remember which words they saw, but they’re also better able to remember what color the taboo words were printed in and even where they appeared on the screen.30

  The upshot is that the taboo Stroop effect could provide further evidence that people are self-monitoring so they don’t accidentally say the wrong word, but it’s also consistent with this alternative, emotion-driven explanation.

  There’s another corroborating effect, similar to Stroop, which comes from a paradigm known as picture-word interference.31 The basic idea is that you have people name pictures of familiar artifacts and organisms—hammers, tigers, and so on. This isn’t hard. But the task gets slightly harder when you print words over the pictures. The words don’t really interfere with your ability to perceive the object, but they can make it harder to name it, depending on how the word and picture are related. The details are tricky, but, generally, if you show people a word that sounds like the name of the picture (for instance, if you print “dock” over a picture of a dog), then people name the picture faster; however, if the word is related in meaning (like “cat” over a dog), people take longer.g