The Secret Life of Pronouns (17 page)

Is it possible to use language analyses to tell which of my letters are real and which ones are self-deceptive? To find out, I analyzed about two hundred letters of recommendation that I had written. To begin, I went back and rated how I truly felt about the student’s potential. This rating was made often several years after writing the letters and was not tainted by the post-writing self-deceptive glow. I then simply correlated the word use in my letters with my ratings. Four reliable patterns emerged.

In my letters for the students I truly felt would do the best, I tended to:

•  
Say more, use longer sentences and bigger words.
The better I felt they would ultimately perform, the more words I wrote. Longer letters reflect the fact that I had more to say about the student. My sentences were also more complex.

•  
Use
fewer
positive emotion words.
This shocked me. The more potential I believed the student had, the
less
likely I was to use words like
excellent
,
nice
,
good
, and
happy
. As I went back to the original letters, the reason became clear. For people best suited for graduate school or a particular job, I tended to give specific examples of their accomplishments. For people not as suited, I tended to wave my hands a bit more and simply say they were great at what they did.

•  
Provide more detailed information.
I noticed that in the stronger letters, I paid more attention to what the students had done rather than just talk about the students themselves. For the promising students, I spent a disproportionate amount of time describing their work and their contribution to whatever projects they worked on. To do this, I used more words that referenced time, space, and motion. Relatively few of my sentences used pronouns like
he
or
she
because I was focusing on the projects as opposed to the people.

•  
Pay little attention to the potential reader of the letter.
Recall that the ways people use pronouns reveals where they are paying attention. Apparently, when writing letters for somewhat weaker students, my mind would float to the reader and away from the student. I included phrases like “as you can see from the candidate’s résumé …” or “I’m sure you will agree that …”

The language analyses indicated that my letters of recommendation said more than I thought they did. I suspect that I’m similar to most earnest letter writers who are not intentionally lying to try to place a student. There is little doubt, however, that some self-deception is at work. The findings point to how I subtly and unconsciously engineer a letter to make a student appear stronger than he or she might be. Do other letter writers employ the same techniques as part of their own self-deceptions? No studies have been conducted on the topic yet. My sense is that there are wide variations in the ways self-deception is expressed. Nevertheless, there appear to be common language markers across the different types of self-deception.

It is easy to imagine creating a system to automatically analyze letters of recommendation. Many companies might be interested in this product. This is both an intriguing and an unsettling idea. For a computerized system to work efficiently, multiple letters of recommendation would need to be collected from each letter writer. It is likely that the computer would do as well or perhaps slightly better than human readers. The real problem is that letters of recommendation are currently not very diagnostic in predicting students’ future performance. A computer program may help in the selection process but the danger is that company executives may start taking the program’s results too seriously. Remember that language analyses are all probabilistic and recommendation letters are often a small part of the decision to hire someone.

THE LANGUAGE OF LYING, LIARS, AND SCOUNDRELS: IN SEARCH OF THE LINGUISTIC LIE DETECTOR

The line between self-deception and all-out deception is not entirely clear. Is the self-deceptive student who claims that the death of his father was not a problem actually trying to deceive his family or friends into thinking he was stronger than he really was? Is the self-deceptive writer of letters of recommendation being deceptive by attempting to influence the people who receive the letters? If there is a linguistic fingerprint for self-deception, is it similar to one that might emerge when people are actively lying to others?

The prospect of a linguistic lie detector has intrigued parents, spouses, teachers, car buyers, people in law enforcement, and just about everyone else for generations. In theory, developing such a system should be possible. Sigmund Freud, for example, suggested that people’s true feelings sometimes leak out and can be seen through subtle speech errors, such as slips of the tongue. Slips of the tongue are most likely to occur when a person is thinking of one thing but having to talk about something else.

Once you start paying attention, slips of the tongue are surprisingly common—in natural conversation, e-mails, online chats, and even formal lectures and speeches. Just last week a friend who works in an office with a domineering boss sent me a message from work: “Are you free for coffee after work? I’d like to have a quit talk with you.” What she meant to say was a “quick talk” rather than a “quit talk.” When we met, I opened our conversation with “So tell me why you want to quit your job.” My friend was amazed at my brilliant psychological insight.

There are many other ways that the words we use can reveal our deceptions. In 1994, Union, South Carolina, residents were horrified to learn that the car of one of their neighbors, Susan Smith, had apparently been hijacked on a nearby country road with her young children still in it. Smith claimed that an African-American male had forced her from the car and had sped away. For over a week, Smith made powerful appeals to the hijacker on national television to return her children. Before she was an official suspect, Smith told the reporters, “My children wanted me. They needed me. And now I can’t help them.”

The FBI agents on the case paid particular attention to that statement. Why had she used past tense when talking about her children? Only someone with knowledge of her own children’s death would talk of them in the past. Several days later Smith confessed to the drowning of her children by driving her car into a lake. Her motive was to get rid of the children so that she could marry a local businessman who didn’t want to be burdened with an instant family.

ON THE SURFACE, building a linguistic lie detector should be simple. Find a large number of instances where people tell the truth and compare their word use with instances where people lie. The problem is that there are dozens of ways that people can deceive. Some are intentional, others are not. Some are large lies with life-changing consequences if the liar is caught, others are no more than parlor games. Some lies involve creating elaborate stories whereas others may simply involve omitting or adding some crucial information to an otherwise true story. Lies also differ in their ultimate goals, which may include gaining status, money, or sex; avoiding shame or punishment; or just having fun. Perhaps most important is the context in which lies are told. A person who is being interrogated by the police is likely to use words differently than the same person who is trying to get an elderly couple to buy some nonexistent ocean-side property. Although the work on linguistic lie detection is in its early stages, some exciting language patterns are beginning to emerge.

CATCHING FALSE STORIES

One reason for Susan Smith’s downfall was that her stories about her missing children didn’t match her language use. Telling a false story is harder than you might think. A convincing lie requires describing events and feelings that you have not directly experienced. In a sense, you don’t “own” a fictional story in the same way you do a true one. Similarly, when constructing a false story you have to make an educated guess about what realistically could have occurred.

Real and Imaginary Traumas
    In the 1990s, Melanie Greenberg, Arthur Stone, and Camille Wortman at the State University of New York at Stony Brook ran a wonderfully creative expressive writing study to see if there were health benefits to writing about imaginary traumas. The authors first enlisted approximately seventy people who reported having had a major traumatic experience in their lives. Half of the people who came to the lab wrote about their own traumas. The essence of each trauma was then summarized in a few lines. Soon, the other half of the participants visited the lab, but this group did not write about their own traumas. Instead, each was given a brief trauma summary from someone who had already written about a trauma. The new group was told to imagine that they had experienced that particular trauma and to write about it as if it had happened to them.

Half of the essays, then, were people’s real traumas and the other half were imagined traumas. Both sets of essays were surprisingly powerful. In fact, the researchers discovered that writing about an imaginary trauma was almost as therapeutic as writing about a real one. In reading the essays it is often impossible to know which ones were real and which ones were made up.

However, computer analyses of the two sets of essays revealed a large number of striking differences. First, people writing about their own real traumas simply wrote more words than people writing about imaginary traumas. They were able to supply more details about what they did and didn’t do. They also used first-person singular pronouns more—words such as
I
,
me
, and
my
. Real stories also included
fewer
emotional words—both positive and negative emotions—than imaginary stories. Finally, the real stories used fewer verbs and cognitive (or thinking) words than the imaginary ones.

This jumble of effects is not that surprising. Consider the meaning of each of the language differences. Writing about real experiences is associated with:

•  
More words, bigger words, more numbers, more details.
If you have experienced a real trauma in your life it is easy for you to describe what happened. You can describe the details of the experience without having to do much thinking. Some of the details include information about time, space, and movement.

•  
Fewer emotion and cognitive words.
If you have lived through a trauma, your emotional state is obvious. For example, if your father died, most people don’t then say “and I was really sad.” It is implicit in the experience. However, people who haven’t experienced the death think to themselves, “Well, if my father died I would feel very sad so I should mention that in my essay.” The person who has had a trauma in the past already has a reasonable story to explain it. The person who is inventing the story must do more thinking—and use more cognitive words to explain it.

•  
Fewer verbs.
There are a number of different types of verbs that can serve different functions in language. When a person uses more verbs it generally tells us that they are referring to more active and dynamic events. For a person who has had a trauma in the past, much of it is over. If you are writing about an imaginary trauma, you are living it as you tell about it. In addition, imaginary traumas cause people to ask themselves, “What would have happened? How would I have felt?” Discrepancy verbs such as
would
,
should
,
could
, and
ought
were used at particularly high rates in the imaginary traumas.

•  
More self-references: I-words.
Recall that I-words signal that people are paying attention to themselves—their feelings, their pain, themselves as social objects. By the same token, the use of first-person singular pronouns implies a sense of ownership. Not surprisingly, people writing about their own traumatic experiences were more acutely aware of their feelings and, at the same time, embraced their traumas as their own.

These language dimensions do a very nice job in distinguishing the real from the imaginary essays the participants wrote. Using some elegant statistics (a cross-validation strategy), we can estimate that our word categories would accurately classify 74 percent of essays as either real or imaginary. This is an interesting case because no deception was involved.

Real Versus Fabricated Stories: The Case of Stephen Glass
    In 1995, the magazine the
New Republic
hired a young and promising writer, Stephen Glass, who had just graduated from the University of Pennsylvania. Within a few months, he began publishing comments and then full-blown articles on a wide range of topics. Between December 1995 and May 1998, Glass published forty-one high-profile and often beautifully written articles. His colleagues and readers often marveled at his ability to find such interesting people to interview and to draw such colorful quotes from them.

Glass, it seems, was able to get such great stories and quotes by just making them up. When he was finally caught by the magazine’s editors, he was fired. An investigation revealed that at least six of his articles were completely invented, another twenty-one were partially fraudulent, and the remaining fourteen were likely trustworthy. An excellent movie on the Glass scandal was eventually made,
Shattered Glass
. And, in an ironic twist of fate, Glass ended up going to law school, presumably in search of truth and justice.

Is it possible that Glass left a linguistic fingerprint on his fabricated stories? Using our computerized text analysis program, I checked them out. By and large, the language analyses found that the completely fabricated and the likely true stories were quite different from each other and the partially fraudulent fell somewhere between the two. Glass’s real or likely true stories were characterized by the use of the following word categories:

•  More words, more details, more numbers

•  Fewer emotion (especially positive emotion) and cognitive words