The idea behind LIWC was that the words people used—whether in a trauma essay or everyday speech—would reflect their feelings and that by the simple process of counting these words we could gain insights into their emotional states. We assumed that angry people would use anger-related words; sad people would use sadness words. In writing about a trauma, the emotional states of our participants should be reflected in their selection of emotionally relevant words.
So, in developing the LIWC program, we created a series of word dictionaries designed to capture different psychological concepts. For example, we built an anger dictionary, now made up of over 180 words, that comprised numerous words related to anger, such as hate, rage, kill, slash, revenge, etc. We also included word stems such as kill so that any word that starts with the letters K-I-L-L, such as killer, killing, kills, and killed, would be counted as well. We then went on to build dictionaries for sadness, anxiety, positive emotions, and other mood states.
The trauma essays differed in multiple dimensions beyond their emotional tone. To cast a fairly broad net we developed other lexicons that measured the occurrence of other types of words, such as the use of different types of pronouns (e.g., first-person singular—such as I, me, and my), articles (a, an, the), different types of thinking-related words that signal cause-effect thinking (cause, because, reason, rationale), and so forth. Before we knew it, we had created almost eighty different dictionaries that we felt would include nearly all of the types of words people commonly use in everyday language.
The reason it took almost three years to get LIWC running was because of the painstaking process of building each dictionary. We employed an army of students who evaluated every word that was part of any dictionary. For example, should the word frustration be included in the anger dictionary? Panels of student judges had to all agree that it was related to anger (in this case, it was).
Thanks to Martha’s programming skills and the thousands of hours spent by our student judges, LIWC was eventually ready to go. The final program instantly analyzed computer-based text or document files and calculated the percentage of words associated with each dictionary. The most recent version of LIWC can analyze thousands of individual digital files in a matter of seconds. Although our initial studies all focused on trauma essays, we eventually moved to poems, novels, blogs, Twitter feeds, letters, IMs, transcripts of conversations, and any other documents that contain words.
To appreciate how a word-counting program works, let’s look at the first two sentences of Lewis Carroll’s Alice’s Adventures in Wonderland:
Alice was beginning to get very tired of sitting by her sister on the bank, and of having nothing to do: once or twice she had peeped into the book her sister was reading, but it had no pictures or conversations in it, “and what is the use of a book,” thought Alice “without pictures or conversation?”
So she was considering in her own mind (as well as she could, for the hot day made her feel very sleepy and stupid), whether the pleasure of making a daisy-chain would be worth the trouble of getting up and picking the daisies, when suddenly a White Rabbit with pink eyes ran close by her.
LIWC would begin its analysis by first counting all the words in the text, which, in this case, is 113. It would then look at each word separately to determine if it was included in any of the existing dictionaries. So, for example, LIWC would first see the word Alice but would find no such word in any of its dictionaries. It would then move to the word was. Voilà! The word was would be in several dictionaries, including the verb dictionary, the auxiliary-verb dictionary, and the past-tense verb dictionary. The count for each of those dictionaries would now be 1. As LIWC proceeded in its task, it would then locate the next word, beginning, in the time dictionary; to in the preposition dictionary; and so forth. Finally, after evaluating all 113 words in the text and assigning each of them to the relevant dictionaries, LIWC would then calculate the percentage of total words that are linked to each dictionary. So, for example, in this passage, about 7 percent of all the words are personal pronouns, 9 percent are articles, and 3.6 percent are words related to emotion.
In analyzing a text, LIWC had many advantages over my troublesome human experts. Programs such as LIWC are 100 percent reliable in that you get the same results every time you run the program on a particular text. They are very fast, able to analyze the collected works of Shakespeare in under twenty seconds. And the results from the analysis of one person’s text can be directly compared with those of anyone else’s.
Despite these admirable features, word counting programs are also remarkably stupid. They can’t detect irony or sarcasm and are singularly lacking in a sense of humor. Particularly damning is that they fail to capture the context of language. One word, for example, can have very different meanings depending on how it is used.
Consider the word mad. The LIWC program counts mad in its anger and negative-emotion dictionaries. If someone said, “I’m mad at you for kissing my new boyfriend,” LIWC’s interpretation of mad as an anger word would no doubt be correct. But if the same person said, “I’m mad about my new boyfriend,” LIWC would be mistaken to sort mad according to its given definitions. In this case, mad means not angry but “crazily happy.” Or, returning to Alice in Wonderland, the Mad Tea Party was not a hostile affair. Rather, mad in this context simply means “peculiar” or “insane.”
LIWC, like almost all word-counting systems, makes lots of errors. It is a probabilistic system. Sometimes it classifies correctly and sometimes it doesn’t. We have now run enough studies to determine that statistically it is usually correct, and the good news is that the more words there are available to analyze, the more precise the system. Another bit of good news is that researchers are developing smarter word-count programs that will eventually take into account syntax, grammar, and context in general.
Given the current state of the art, it is little wonder that serious linguists and literary scholars find word-counting programs somewhat distasteful. The thing is, linguists care about language and literary scholars care about literature. And, in my own begrudging way, I care about these things too. But what I’m really interested in is the relationship between word use and people’s psychological states. Can we identify features of language that reveal how people are thinking? And, if so, can we use this information to change their thinking in a beneficial way?
IDENTIFYING HEALTHY WRITING WITH A COMPUTER PROGRAM
The entire purpose of developing LIWC was to see if the ways people wrote about their traumatic experiences could predict later improvements in their health. Put another way, could we use LIWC to identify healthy writing?
During LIWC’s development, several expressive writing studies had been conducted that could answer this question. In my lab, three had been conducted with college students. Others had been run by colleagues who had relied on more diverse samples, including a study of maximum-security prisoners, a group of New Zealand medical students, and a cohort of senior engineers recently laid off from their jobs. Now that we had essays from a wide range of people across six different studies, we could get our computer program to find which word categories were associated with healthy writing. Three important findings emerged.
THE IMPORTANCE OF POSITIVE EMOTIONS
A rough measure of people’s emotional state can be found by counting words in their trauma essays that signify positive emotion (e.g., love, care, happy) and negative emotion (e.g., sad, pain, anger). The results from the six writing studies were somewhat unexpected. Overall, the more people used positive emotions while writing about emotional upheavals, the more their physical and mental health improved in the weeks and months after the experiment.
Negative emotion words showed a different pattern. People whose physical health improved the most from writing used a moderate number of negative emotion words. That is, people who expressed negative emotion language at very high rates did not benefit from writing—almost as if they were awash in their unhappiness. By the same tok
en, those who used very few negative emotion words did not benefit—perhaps a sign that they were not acknowledging the emotional impact of their topic. The emotional findings, then, suggest that to gain the most benefit from writing about life’s traumas, acknowledge the negative but celebrate the positive.
THE IMPORTANCE OF CONSTRUCTING A STORY
One of the exciting aspects of the LIWC program was that we were able to identify word categories that reflected the degree to which people were actively thinking. Two of the cognitive dimensions included insight or self-reflection words (such as think, realize, believe) and another made up of causal words (such as because, effect, rationale). The people whose health improved the most started out using fairly low rates of cognitive words but increased in their use over the four days of writing. It wasn’t the level of cognitive words that was important but the increase from the first to last day. In some ways, use of insight and causal words was necessary for people to construct a coherent story of their trauma. On the first writing session, people would often spill out their experience in a disorganized way. However, as they wrote about it day after day, they began to make sense of it. This greater understanding was partially reflected in the ways they used cognitive words.
These findings suggested that having a coherent story to explain a painful experience was not necessarily as useful as constructing a coherent story. This helped to explain a personal observation that had bothered me for years. When the first writing studies were published, my work was often featured in the media. At cocktail parties or informal gatherings, I sometimes found myself to be a trauma magnet. People who knew about my research would gravitate to me in order to tell me all about their horrific life experiences. Many of them also were in very poor physical health. At first, I thought that their talking about their stories would be good for them. However, I’d see the same people at another gathering months later and they would often tell me exactly the same stories and their health would be unchanged.
The word count research revealed the problem. The people telling their traumatic stories were essentially telling the same stories over and over. There was no change to the stories, no growth, no increase in understanding. Repeating the same story in the same way is not unlike ruminative thinking—a classic symptom of depression.
There is an important lesson here. If haunted by an emotional upheaval in your life, try writing about it or sharing the experience with others. However, if you catch yourself telling exactly the same story over and over in order to get past your distress, rethink your strategy. Try writing or talking about your trauma in a completely different way. How would a more detached narrator describe what happened? What other ways of explaining the event might exist? If you’re successful, research studies suggest that you will sleep better, experience better physical health, and notice yourself feeling happier and less overcome by your upheaval.
THE IMPORTANCE OF CHANGING PERSPECTIVES
Thanks to the LIWC program, we found that three aspects of emotional writing predicted improvements in people’s physical and mental health: accentuating the positive parts of an upheaval, acknowledging the negative parts, and constructing a story over the days of writing. More complex analyses soon revealed another dimension of word use that no one had seen before.
In the 1990s, a group of researchers at the University of Colorado introduced a computer program called Latent Semantic Analysis (or LSA) that could track patterns of word use both within and across different essays. The beauty of LSA was that it could mathematically compare how similar any two writing samples were. On the surface, then, we could determine if people who wrote about the same topics from day to day might benefit more than those who wrote about different topics. It was a grand idea, but try as we might, we couldn’t find any good evidence to support it.
One of my graduate students, Sherlock Campbell, had spent almost a year getting the LSA program to work. The more he and I thought about the LSA project, the more we realized that we had been thinking about language the wrong way. Instead of analyzing the content of what people were writing, why not analyze their language style? To do this, we needed to turn the LSA program on its head. Instead of analyzing the content of the essays by focusing on nouns, regular verbs, and adjectives, we asked the program to focus on the words that revealed writing style. Writing style, we were learning, was generally revealed through function words, including pronouns, prepositions, articles, and a small number of similar short but common words.
The results were breathtaking. (OK, if you are not a computational linguist, “breathtaking” may be a bit of an overstatement. You had to be there.) The more people changed in the ways they used function words from writing to writing, the more their health later improved. As we started to focus on different classes of function words, one particular group of culprits stood out as more important than the others: personal pronouns. More specifically, the more people changed in their use of first-person singular pronouns (e.g., I, me, my) compared with other pronouns (e.g., we, you, she, they), the better their health later became. The effects were large and held up for study after study.
After spending over a year on the computer program, Sherlock was thrilled. He took great pride in noting that we had discovered the “secret life of pronouns.” And he deserves the credit for the title of this book.
The findings may sound esoteric but in real life they aren’t. The writings of those whose health improved showed a high rate of the use of I-words on one occasion and then high rates of the use of other pronouns on the next occasion, and then switching back and forth in subsequent writings. In other words, healthy people say something about their own thoughts and feelings in one instance and then explore what is happening with other people before writing about themselves again.
This perspective switching is actually quite common in psychotherapy. If a man visits his therapist and begins repeatedly complaining about his wife’s behavior, what she says, how aloof she is, and so forth, the therapist will likely stop the client after several minutes and say, “You’ve been talking about your wife at length but you haven’t said anything about yourself. How do you feel when this happens?” Similarly, if another client—a woman in this case—with marital problems sees her therapist and spends most of her time talking about her own thoughts, feelings, and behaviors without ever talking about her spouse, the therapist will probably redirect the conversation in a similar way by asking, “You’ve told me a lot about your own feelings when this happens—how do you think your husband feels about this?” Perhaps like good therapy, healthy writing may involve looking at a problem from multiple perspectives.
WORDS AS MIRRORS, WORDS AS TOOLS
Stand back for a minute and consider the meaning of all of our findings regarding expressive writing. Writing about emotional upheavals can improve people’s mental and physical health. Not all people benefit from this exercise however. Those who do benefit tend to write differently from those who don’t. Healthy writing involves positive emotion words, a moderate use of negative emotion words, increasing use of cognitive words, and changes in pronoun use. Translating these effects into everyday language: People who benefit from writing express more optimism, acknowledge negative events, are constructing a meaningful story of their experience, and have the ability to change perspectives as they write.
Most surprising, though, was that these discoveries were reflected through people’s use of a small number of almost-invisible stealth words. The stealth words, which had been there all along, reflected critical changes in the ways people were thinking.
These language findings are certainly interesting, but can we put them to good use? If we bring people into the lab and encourage them to use positive emotion words, increase their use of cognitive words, and oscillate in their use of personal pronouns while they write, will their health improve? In other words, do words reflect a psychological state or do they cause it?
Over the years, several studies have been conducted to try to answe
r this question. In one elaborate experiment, Cheryl Hughes, a former student of mine at Southern Methodist University, gave different students lists of words that she asked them to use in their expressive writing. Some received lists of positive emotion words, others received negative emotion words, some were given cognitive word lists and others weren’t. While she succeeded in manipulating the words people used in the predicted directions, the writing had no effect on health. Other clever attempts have been made to get people to change the rates at which they use cognitive words or to change the types of pronouns from writing to writing while addressing emotional topics. The current evidence is convincing: Word use generally reflects psychological state rather than influences or causes it.
That words we use mirror our thoughts and feelings is not a startling revelation. But the findings point to ways we can now use word analyses to change people’s thinking. Recall that healthy writing is characterized by an increasing use of words such as because, cause, effect, reason, and related cognitive words. Simply requiring people to use the words at higher rates over the course of writing has no meaningful effects—the writers are simply focusing on words and not their underlying purpose. However, if we encourage people to write about a trauma and to work to construct a meaningful story, their writing takes on a more dynamic tone. They begin to stand back and look at their trauma with a broader perspective. The cognitive work they put into the story results in a better product and one that is more likely to allow them to get through their trauma.
The Secret Life of Pronouns: What Our Words Say About Us Page 2
