Are There Hidden Messages in Pronouns?

James Pennebaker says computers reveal secret patterns.

James Pennebaker

Some 110 years after the publication of the Psychopathology of Everyday Life, in which Sigmund Freud analyzed seemingly trivial slips of the tongue, it’s become common knowledge that we disclose more about ourselves in conversation—about our true feelings, or our unconscious feelings—than we strictly intend. Freud focused on errors, but correct sentences can betray us, too. We all have our signature tics. We may describe boring people as “nice” or those we dislike as “weird.” We may use archaisms if we’re trying to seem smart, or slang if we’d prefer to seem cool. Every time we open our mouths we send out coded, supplementary messages about our frame of mind.

Although much of this information is easy to decode (“nice” for “boring” won’t fool anyone), linguistic psychologist James Pennebaker suggests in The Secret Life of Pronounsthat lots of data remain hidden from even the most astute human observers. “Nice” and “weird” are both content words; he’s concerned with function words such as pronouns (I, you, they), articles (a, an, the), prepositions (to, for, of), and auxiliary verbs (is, am, have). We hardly notice these bolts of speech because we encounter them so frequently. With the help of computer programs to count and scrutinize them, however, patterns emerge.

Sounds enticing; sounds, in fact, rather like a publisher’s fantasy pitch, combining the strangely long-lasting craze for language books laced with pop psychology, and the added hook, the modern touch, of a computer that observes and catalogs beyond measly human capacity: a Watson for the psychiatric establishment. To Pennebaker’s credit, his claims are fairly modest, especially when compared with those of Deborah Tannen and other practitioners of the word-sleuth genre. (He doesn’t promise that if we change our pronoun usage we’ll see tangible improvements in our social lives.) The problem is that much of what he turns up is even more modest than he seems to notice. Counting function words as they’re used in ordinary life often yields the opposite of what Freud detected in confessions from the couch: confirmation of the obvious.

The most ingenious application Pennebaker proposes for function-word analysis is lie-detection, something of a dark art. Several years ago, Pennebaker and a couple of colleagues recruited 200 students and asked them to write two essays about abortion, one espousing a true belief, the other a falsehood. They asked another group to state their true and false takes in front of a video camera. When judges were called in to figure out which was which, they were accurate 52 percent of the time. (50 percent is chance.) A computer, programmed to look for specific “markers of honesty” gleaned from previous studies, performed much better, with a 67 percent accuracy rate. Truth-tellers, Pennebaker explains, tend to use more words, bigger words, more complex sentences, more exclusive words (except, but, without, as in the sentence “I think this but not that”), and more I-words (I, me, my, etc.). Liars, apparently, trade in simple, straightforward statements lacking in specificity because—Pennebaker posits—it’s actually pretty difficult to make stuff up. They avoid self-reference because they don’t feel ownership of their expressed views.

When Pennebaker dips into the more general field of “emotion detection” (he calls it that), his word-counting feels a bit Rube Goldberg-ish. After Sept. 11, 2001, Pennebaker and a colleague saved the postings of over a thousand amateur bloggers. They found that “bloggers immediately dropped in their use of I-words” following the attacks, and that their use of we-words almost doubled. Pennebaker takes these fluctuations to mean that “shared traumas bring people together,” “shared traumas deflect attention away from the self,” and that “shared traumas, in many ways, are positive experiences” (because people feel more socially connected). The brute fact that Sept. 11 influenced pronoun usage may interest readers, but Pennebaker’s analysis merely reiterates long-held psychological dogma. (Try Googling “shared traumas bring people together.”) I can’t help but wonder if Pennebaker—albeit unconsciously—interpreted his results to match the conventional wisdom.

Perhaps that’s harsh: Certainly there’s nothing wrong with devising yet another way to elucidate common human responses, and Pennebaker’s experiments are always imaginative. Yet it’s often the case that his conclusions, especially the ones he draws from I-word usage, are heavily dependent on context and prior knowledge.

In one chapter, Pennebaker notes that Rudolph Giuliani demonstrated a dramatic increase in I-words during the late spring of 2000, when he was still mayor of New York. Pennebaker fills us in that “Giuliani’s life [was] turned upside down. … He was diagnosed with prostate cancer, withdrew from the senate race against Hillary Clinton, separated from his wife on national television … and, a few days later, acknowledged his ‘special friendship’ with Judith Nathan.” Pennebaker adds that “by early June, friends, acquaintances, old enemies, and members of the press all noticed that Giuliani seemed more genuine, humble, and warm.” So it’s reasonable to conclude that Giuliani’s ascending I-word usage reflected a “personality switch from cold and distanced to someone who [due to a few significant setbacks] was more warm and immediate.”

But we already knew that. If we didn’t, where would Pennebaker’s method leave us? He argues, at various points, that the following groups use I-words at higher rates:          

1.      Women
2.      Followers (not leaders)
3.      Truth-tellers (not liars)
4.      Young
5.      Poor
6.      Depressed
7.      Afraid (but not angry)
8.      Sick

The common thread unifying these seemingly random clusters is, roughly, an enhanced focus on personal experience. Sick and depressed people dwell on their conditions and are thus more likely than their healthy counterparts to talk about themselves. Followers, in conversation with leaders, might be after something: “I was wondering if I could have a raise.” That’s pretty close to a tautology, though, and does nothing to solve the problem that, without insider information, it’s impossible to know which condition or attribute I-usage reflects. A word-count-wannabe presented with Giuliani’s speeches might deduce, erroneously, that the mayor had become more truthful, or less leaderly, or had lost money.

For obvious reasons, I’m unusually attuned to my pronoun usage at the moment, and I’ve noticed a thing or two. I start off this essay with lots of we-words (16 in the introduction), and sprinkle them throughout. With the exception of the section you’re currently reading, I drop only one self-referencing I (in the fifth paragraph). I don’t deny that this imbalance might mean something. Perhaps it indicates that, like politicians who drone on about what “we” expect from the president, or how “we” want a return to old-fashioned American values, I’m trying to imply audience agreement when, in truth, I have no clue what the audience thinks. But you don’t need to count pronouns to figure that out. You only need to know that you’re reading a book review.