The Secret Life of Pronouns: What Our Words Say About Us

by James W. Pennebaker

  The Wikipedia project hints that it might be possible to assess the connections between people by simply measuring how people in a community use language. Another amazing Internet development has been the growth of the site Craigslist.org. In most U.S. cities, Craigslist allows people to buy, sell, and give away a large number of goods and services. The ads themselves are free and people can write as much as they like and even include pictures. The demise of many American print newspapers has been attributed, in part, to Craigslist in that people are more likely to sell their cars, rent their houses, or give away their puppies on Craigslist than use paid classified advertisements in their newspaper.

  In 2008, my students and I saved copies of all Craigslist ads in thirty different midsize communities across the United States. Because there are so many categories of ads, we only looked at ads for cars, furniture, and roommates. Within each category, we analyzed between six thousand and ten thousand ads and evaluated how people in the different cities used function words. More specifically, we were curious how similarly different communities used pronouns, prepositions, articles, etc.

  The results were fascinating. In several of the communities, people tended to write their ads using the same language style as their neighbors. In Portland, people generally wrote in a personal style with a slight negative emotional tone. People in Salt Lake City, on the other hand, were exceptionally upbeat in their ads. Although Craigslist writers in Portland and Salt Lake City may have had distinctive styles, within each city, people were thinking about their ads in the same ways. In contrast, residents of Bakersfield, California, and Greensboro, North Carolina, were much more scattershot in their writing styles—meaning some people may have been very formal in their writing and others very informal. They simply didn’t speak with one voice.

  Using a method similar to language style matching, we were able to calculate the degree to which residents in any given town used language in similar ways. The findings suggested that the more similar the community’s use of language, the more cohesive the city.

  As you look at the table that lists the ten most and least linguistically cohesive communities, the difference between them may not be immediately obvious. It’s not a matter of north versus south, liberal versus conservative, rich versus poor, or even a function of the towns’ racial makeup, age distribution, or migration into or out of the city. Rather, one of the most striking differences between the two lists is that the most cohesive communities have higher and more equal income distributions than the least cohesive. Using a statistic called a Gini coefficient, demographers can determine the degree to which wealth is spread around in any given city, state, or country. Cities where the residents use language in similar ways tend to be communities that are more similar in terms of their incomes. The bigger the split between the rich and poor of a community, the more varied their writing styles for Craigslist ads.

  RANKING OF COMMUNITIES BY LANGUAGE STYLE MATCHING STATISTICS FOR CRAIGSLIST ADS

  MOST LINGUISTICALLY COHESIVE (TOP 10)

  LEAST LINGUISTICALLY COHESIVE (BOTTOM 10)

  Portland, Oregon

  Bakersfield, California

  Salt Lake City, Utah

  Greensboro, North Carolina

  Raleigh, North Carolina

  Louisville, Kentucky

  Birmingham, Alabama

  Oklahoma City, Oklahoma

  Rochester, New York

  Dayton, Ohio

  Hartford, Connecticut

  El Paso, Texas

  New Orleans, Louisiana

  Jacksonville, Florida

  Richmond, Virginia

  Columbia, South Carolina

  Worcester, Massachusetts

  Tulsa, Oklahoma

  Tucson, Arizona

  Albany, New York

  Note: Cohesiveness is calculated by the degree to which people in the various communities used function words at comparable levels in their Craigslist ads.

  You can see why the Gini statistic is important. In communities where there is a much larger split between the rich and the poor, it’s less likely that the different elements of the city will interact. As the range of income within a town becomes smaller, the residents will likely have more in common with one another and should talk more with others. Not coincidentally, they should also have more objects and services that others in the community would want to exchange on Craigslist.

  The magic of this project is not about the links between income distributions and social patterns in cities. Rather, it shows how words in the most mundane of places can reveal important information about a community’s social ties. All groups, whether families, work groups, companies, or entire cities, leave trails of their social and psychological lives behind in the words their members use in communicating with each other. Words are one of the human-made elements that connect our thoughts and ideas across people. By tracking our words, we get a sense of the social fabric.

  GROUP SPOTTING: HOW WORDS REVEAL WHAT A GROUP IS DOING AND WHERE IT IS

  Consider what we know so far. Individuals use words—especially pronouns—to signify their sense of belongingness to a wide range of ever-shifting groups. Whereas we-ness measures reflect feelings of group identity, synchrony measures, such as language style matching, tap a shared worldview by the group members.

  Taken together, the various findings point to the shared mind-set of most groups. Within any group—whether a family, school, work team, or even a community—people soon start talking alike. In small groups, they discuss the same topics with each other, and in a sense, it isn’t too shocking that their function words are similar. But as groups get larger and not everyone necessarily talks with everyone else, it is impressive that the group still maintains its own linguistic style.

  Sociolinguists would not be surprised that large groups share similar linguistic styles. There is now a generation of important research tracking how neighborhoods, cities, and entire regions of a country begin to develop their own unique accents. Further, people can usually modulate their accents depending on who they are talking with. If my wife gets a phone call at home, I can usually make a reasonable guess who she is talking to by her cadence, volume, tone, and even accent. If my brain were wired differently, I could probably also hear her function word use and make even better guesses.

  Think about the implications of the “contagion” of function word use. If function words really work a bit like accents, our language should vary depending on the group we are in, what the group is likely doing, and perhaps even where the group resides. We may not be aware of it but we likely shift in our use of function words depending on what groups we happen to be in and what the groups happen to be doing.

  Taken together, every group has its own linguistic fingerprint. Parts of the fingerprint may reveal how well the group works together and the degree to which its members identify with the group. Other parts of the fingerprint reflect how the group thinks and feels and how emotionally open, formal, detached, and so forth it is. In theory, the language from any group gathering should provide clues to the dynamics of the group and its members.

  WORD CATCHING IN PUBLIC PLACES: THE SOCIAL ECOLOGY OF WORDS

  Groups, of course, have different purposes. Sometimes, we are in groups that work on a task, play a game, or just socialize. Different tasks demand different ways of speaking. By analyzing the words that different group tasks demand, we can get a new perspective on the nature of the groups themselves.

  Several years ago, my students and I carried digital recorders to various public places. We would slowly walk through restaurants, crowded hallways, post offices, college basketball and young children’s soccer games, classrooms, and poolrooms with our recorders held high. The “word catching” project had very few rules other than we had to keep moving the entire time and not hang around any given conversation for more than a few seconds. The goal was simply to catch words in different types of locations. Over the years, we collected additional language samples that included
people working in groups and places where strangers were meeting for the first time. One by one, every language sample was transcribed and added to our Word Catcher archive.

  “Why bother doing such a screwy project?” many of my friends and colleagues would often ask. I wanted to know if and how people changed the ways they used words depending on the situations they were in. The results were clear: Every context has its own linguistic fingerprint.

  Spend a minute and ponder the findings in the “Language of Situations” table. Every general situation is associated with a very different word pattern. When people are seated and talking with others over lunch or dinner, or even seated in a park, they use high rates of I-words, he-she words, past tense, and negations and, at the same time, very low rates of articles. This may sound obscure but it’s not. These word patterns suggest storytelling. And, in fact, reading the transcripts, it is obvious that people who are seated and talking with others are telling stories about themselves and others.

  In sports settings, the language is more upbeat and lacking in introspection. Whether playing billiards or watching a sporting event, both men and women are immersed in the game and not focused on themselves. They are living in the moment while feeling as though they are part of a group. This is probably the allure of sports—they serve as an escape from the self. There is an interesting parallel with the language of work groups. They, too, use high rates of we-words and low rates of I-words. The big difference is that work groups use far more complex language and tend to express more negative feelings.

  People who are talking with others while walking around are personal, upbeat, and superficial. “Hey, how ya doing? I’m going to Serena’s later. Things are going good. Later, dude.” No deep analytic thinking here—just maintaining a friendly public face. It’s interesting to compare the superficial people in motion with strangers who happen to be waiting for an experiment or are thrown together in a room to simply get to know each other. Like people in motion, strangers talking to each other tend to be upbeat. They are different in being less personal and more vague in their claims.

  These same language patterns appear for online chat rooms as well. Chat rooms for strangers trying to meet others use words like our stranger groups. People in chat rooms devoted to sports talk like real people at real sporting events. And chat rooms made up of people who know each other have conversations that use words similar to those among friends at dinner.

  To get a sense of how different the language of these contexts is, imagine that we gave our computer hundreds of new transcripts from recordings across similar situations. How well could the computer categorize which language went with which situation? In this example, there are five different contexts or situations, meaning that a computer should correctly assign language samples to the right category 20 percent of the time by chance alone. Our computers do much better, averaging 84 percent correct.

  Think about the logic of these findings. The situations we are in define the words we use. Imagine spending a weekend with your friends. If you go to a sporting event, you will use one set of words; you then go to a bar and you call forth a different group of words. All of a sudden, you find yourselves in a grocery store and yet another pattern of words is exchanged. The effects are so strong and predictable, one could even claim that the situations you are in demand that you talk in a certain way. OK, “demand” is a little strong. Nevertheless, where we are and what we are doing primes us to think and talk in highly specific ways.

  On the surface, this may sound trivial. When playing or watching sports, people tend to talk about the game. When eating dinner with friends, most of us talk about shared friends or experiences from the past. Contexts or situations dictate the topics of conversation, which undoubtedly influence people’s use of stealth or function words. But there is more to it. Function words tell us the ways that groups and the people who inhabit them are paying attention to their worlds, relating to others, thinking, and feeling.

  The word-catching studies also provide an interesting and almost upside-down way of thinking about stealth words. Most of this chapter and book has been devoted to showing how the words people use reflect who they are. These same words can also tell us where people are and what they are doing. Feed the transcript of a conversation into the right computer and it will tell you about the individuals in the conversation, their relationships, and their situation.

  TRACKING THE GEOGRAPHICAL LOCATION OF GROUPS

  This is where things start to get a little creepy. If groups of people tend to talk and write in similar ways, it should be possible to estimate the physical location of the groups. This goes beyond estimating if someone is at the mall, in a park, or eating in a restaurant. By using both content and function words, it is possible to make educated guesses about which particular mall, park, or restaurant the speakers happen to be visiting.

  The logic of linguistic tracking is similar to that of identifying accents. Depending on what country you live in, you probably do better than chance in picking out which region a speaker hails from. In the United States, the southern drawl, the midwestern twang, and the New York dialect can often be spotted within a few seconds by listeners. Computerized text analysis programs, of course, don’t analyze accents but instead focus on specific words or word patterns that may be common to different geographic regions.

  Regional differences in the naming of things exist for a large array of foods, objects, and behaviors. If you want a soft drink in the Northeast, you will likely ask for a soda, in the South you will ask for a Coke (meaning any type of soft drink—not just Coca-Cola), and in the Midwest, a pop. Depending on where you live, you might refer to that green leafy herb in salsa or Vietnamese pho as cilantro, Mexican parsley, Chinese parsley, or coriander. If you don’t want the truck in front of you to splash mud on your front windshield, you better hope that it is equipped with mud flaps (in the West), splash guards (in the Midwest), or splash aprons (in the East). And if you are driving in England, hope it has mud flaps (two words), or in Australia, mudflaps (one word). Later, when you go for a row in your canoe, be careful because you don’t want it to tip (U.S. East and West) or tump (U.S. South) over.

  One can imagine that by cataloging regional names, we could begin to isolate where people are from. All that would be needed would be to wait until speakers or writers mention their words for soft drinks, spices, or how their boats rolled over. Instead of waiting for relatively obscure words or topics to emerge, it might be more efficient to track more common words. Function words, perhaps?

  AS WE’VE SEEN, stealth words vary by context. It also happens that they can vary according to the geographical regions. Some regional differences in function word use are well known. For example, if people are talking to several of their friends, they might refer to them as “all of you” if from Seattle, “youse guys” if from New York, “y’all” if from the American South, or even “yat” in Louisiana (as in “Where are you at?”).

  Even in relatively formal writing, regional differences exist in function word use. Cindy Chung and I had the opportunity to test this idea using thousands of essays people had written in response to the nationally syndicated radio program This I Believe. The original version of This I Believe ran briefly in the 1950s with journalist and radio commentator Edward R. Murrow. Over the course of a couple of years, Murrow invited some of the cultural leaders of the time to summarize their most important beliefs. The weekly broadcasts of politicians, sports stars, scientists, philosophers, and others captured the imagination of radio listeners. In 2005, Jay Allison and Dan Gediman resurrected the idea for National Public Radio. Rather than relying on celebrities, regular listeners were encouraged to contribute their own This I Believe essays. Over a four-year period, over seventy thousand essays were submitted, although only about two hundred were ever read on air. Most of the essays were eventually posted online for the public to read (www.thisibelieve.org).

  Working with Allison and Gediman, we analyzed about 37,500 es
says. A number of the stories were riveting, some tragic, many funny, most touching and inspirational. Regional differences emerged in terms of the topics of the essays themselves. Stories about sports were most common in the Midwest, racial issues in the South, and science in the Northeast.

  Relative use of words within an immediacy cluster (I-words, short words, present-tense verbs, nonuse of articles) and making-distinction cluster (exclusive words, negations, causal words, nonuse of inclusive words). Darker regions reflect higher usage. Language samples are based on about 37,500 U.S. This I Believe essays.

  Not only did people differ in their topics, but they also differed in terms of their function words. As you can see on the top map, people in the middle of the country tended to use the highest rates of I-words, present-tense verbs, and short words. As you might recall from earlier chapters, this constellation of words reflects psychological immediacy, wherein writers tend to be in the here and now. The darker areas reflect higher rates of immediacy. Contributors from the Northeast and the West were the least personal and social in their writing and the most concrete. These low levels of immediacy reflect a language style of psychological distance and formality.

  Recall from earlier chapters that function words often clump together in a way that reflects analytic thinking. People who think analytically often make distinctions between ideas. In making distinctions, it is necessary to use words such as conjunctions (but, if, or), negations (no, not), and prepositions (with, over). In the bottom map, people in the middle of the country make the most distinctions and those in the far Northeast make the fewest.