Verified by 
All human societies use language. As far as we know, they always have. Over the course of history, humans have written nearly 130 million books, containing over half a trillion words. The average person will speak over 860 million words in their lifetime, and current-day humans will use language to communicate with over 80,000 people over the course of a lifetime.
Consider how difficult it would be to know anything about history—or our own former lives—if it weren’t for language. Without written texts, we wouldn’t know about the empires of Babylon, Rome, China, or Maya. Without letters and speeches, events like major wars, economic depressions, and even our own relationships would recede and eventually disappear from memory. It’s hardly surprising, then, that analysis of written and verbal text is at the bedrock of history, anthropology, psychology, sociology, and any other science that deals with human development over time.
At the turn of the century, however, something changed in the way people studied language: it became possible to digitally study written and verbal text from hundreds of years ago. Software emerged that could quickly and efficiently track how often certain words and phrases appeared across millions of books. Computer science algorithms materialized that could identify groups of words and phrases based on whether people tended to use them together. In just a few clicks, it was possible to illustrate how people were using language over long stretches of time, and these illustrations seemed to capture previously unnoticed episodes of cultural change.
These technological advances sparked an exciting wave of research on language and culture. For example, a team of researchers from the University of Michigan and the University of Texas, Austin found that catastrophic events such as the 9/11 terrorist attacks led people to use more social and communal language. The cultural psychologist, Patricia Greenfield, used a Google tool called the “N-Gram viewer” to show that Americans had been using more “individualistic” language over time—including words like “self,” “unique,” and “individual”—at the expense of collectivistic language such as “obedience,” authority,” and “belong” (I’ve reproduced these trends below using Google’s N-Gram Viewer). Two other cultural psychologists named Igor Grossman and Michael Varnum built on this work by showing that rising socioeconomic status was partly responsible for growing individualism.
I have also done some of this research. Alongside my collaborator Dr. Michele Gelfand and a computer science colleague, Soham De, who now works at Google, I discovered that people seemed to be using more “loose” words over the past 200 years that indicated a positive attitude towards freedom (e.g. “allow,” “autonomy”) while using fewer “tight” words that indicated a negative attitude towards freedom (e.g. “restrain,” “constrain”).
This 200-year “loosening” of American culture was closely linked to some intriguing behavioral trends. As people began using increasingly more freedom-positive words, they also seemed to exercise less self-restraint. Periods with frequent “loose” language had the highest rates of household debt, high school truancy, and adolescent pregnancy. But freedom-positive words were also linked to innovation. Periods with frequent “loose” language also had the highest rates of feature film production, patent application, trademark application, and even had the highest rates of uncommon and creative baby names. We recently published these findings in the journal Nature Human Behavior.
If we had done this research 20 years ago, it would have taken us years. We would have needed to painstakingly read and analyze hundreds of books, and our results would have been nowhere near as persuasive or comprehensive as our 21st-century digital approach. The same is true for every other study that has used a computer-driven approach to study cultural change through language use.
But there is a tradeoff to this digital approach, and it can be difficult to truly know whether culture is changing when people begin to use language differently. Do increases in the frequency of words like “allow” and “autonomy” really mean that culture is becoming looser? Do decreases in words like “belong” really mean that culture is becoming more individualistic?
The Trouble With Language
For a system designed for communication, language can be very misleading. Consider the joke about two hunters walking through the woods when one of them seizes up and collapses. The other hunter assumes his friend is dead and calls 911 in a panic. When the operator tells him to calm down and make sure his friend is really dead, she hears a gunshot and then the hunter’s voice asking “done, now what?” History is littered with similar mix-ups. Some accounts suggest that the “Yucatan Peninsula” in Mexico got its name because “Yucatan” is the Mayan word for “I don’t understand what you’re saying.” When Spanish explorers asked what the region was called, Mayans told them “Yucatan” because they did not understand the question.
Consider all the misunderstandings that could arise from digital analyses of language change. Just to name a few examples: (1) decreases in word usage over time may just mean that the word is being replaced (English speakers now say “before” rather than “ere”), (2) words could be used in ways that reverses their meaning (“society does not ‘allow’ people to be free”), and (3) increases in word usage over time could signal a literary trend that has nothing to do with the word’s original meaning (when people say “Netflix and chill,” they aren’t talking about the temperature).
Many studies and commentaries have pointed out these limitations. A well-known article by a University of Vermont research group summarized them in an article called “Characterizing the Google books corpus: strong limits to inferences of socio-cultural and linguistic evolution.” A more recent article by New Scientist titled “Can scanning books really reveal if the US is becoming more tolerant” raised some of these points again in a thoughtful critique of my own research. These articles make good points and leave us wondering what change in language really mean, and if it really has anything to do with culture.
Some Potential Solutions
These limitations bring me to this article’s eponymous question: “When language changes, is a culture changing?” My short answer is “yes, but not always.” My longer answer is that research on language and culture is very tricky, but there are three key strategies that researchers can use to identify trustworthy findings. I recommend looking for these strategies when you read research that uses language to understand cultural change. And I recommend using these strategies if you are thinking of doing this research.
Strategy 1: Look at more than just a few words. Pretend that you are a scholar doing research on fruit and American society. You wonder whether people discuss fruit more or less than they used to, and you use a digital approach to identify a curious trend. People wrote the word “apple” less and less frequently until about 1980, and then the word began to rebound, trending upwards in usage until today. You might conclude that fruit is experiencing a resurgence in American culture, even though Apple Inc. is probably responsible for the uptick of “apple” usage.
This error points to the need for researchers to use many related words when they do research on language and cultural change. In the example above, someone who used the words of 50 different fruit would probably discover more reliable results than someone just looking for usage of the word “apple.” In our research on freedom and constraint, we used 20 different “loose” words and 20 different “tight” words to identify trends over time. Moreover, we made sure that these words were rising together and falling together in their usage over time, which suggested that upticks and drops in the usage of these words were meaningful. If changes in word usage simply reflected negations, idiosyncratic expressions, or word replacement, these words would not have co-occurred so strongly across history.
Strategy 2: Make sure language can predict non-linguistic trends. When humans face uncertainty, we usually trust our senses to show us the way. We smell milk to make sure it hasn’t gone bad. We try keys on a chain one at a time to find the one that unlocks a new door. In much the same way, researchers should use common sense to test if language trends mean what we think they mean. If a researcher thinks that the usage of “apple” reflects American fruit consumption, they should correlate “apple” usage in a year with the number of apples sold in that same year.
Sometimes these other measures of cultural change are not available or are only available for short periods of time, but often there is enough to get a good idea of whether linguistic trends really measure cultural change. For example, in our paper on freedom and constraint, we found that years where people used “loose” words appeared to have lower rates of societal regulation: fewer laws passed, fewer Supreme Court cases heard, lower religious engagement, and more profanity in film and television. In Igor Grossman and Michael Varnum’s research, they found that increases in individualistic language corresponded to higher divorce rates, smaller family sizes, and a growing number of people living alone. These checks are vital to making sure that language change is tapping real fluctuations in society.
Strategy 3: Don’t pick words yourself. When starting a study of language use and cultural change, it is tempting to pick words that seems valid measures on the surface and begin your analyses. It seems clear that “self” is a good linguistic proxy of individualism and “autonomy” is a good linguistic proxy for freedom. But this strategy can be dangerous. It is tempting, for example, to only choose words that appear to follow the pattern that you expect to find. Perhaps you have a hunch that American society has grown more individualistic: it would be easy to simply choose words related to individualism that appear to be growing more frequent over time while excluding words that appear to be growing less frequent. You may also choose words that are familiar to you but that no one else uses very often. Researchers might commonly use a word in their questionnaires that nobody uses in real life.
The best way to overcome these problems is to avoid choosing the words in your analysis. In our paper on freedom in American society, for example, we used a computer algorithm that picked out words people appeared to use together when discussing freedom related topics. The algorithm we used, which is named word2vec, can estimate the semantic similarity of any two words by posting words in a massive multi-dimensional space. The more often people use words together, the closer they will be in this space.
This is perhaps the most difficult of the three strategies to adopt. As scientists of cultural change, we are supposedly experts in this domain, and so it seems natural for us to select the words in our studies. But in combination with the other two strategies, outsourcing the choice of words that go into a study makes for a more scientifically rigorous approach when using language to analyze cultural change.
Closing Thoughts
We live in an exciting time to study cultural change. Computer-driven approaches mean that we can instantly detect and illustrate fascinating shifts in how people use language, with clear implications for tracking changes in society. With these approaches, we can answer questions that previously would have been impossible to even conceive of. For instance, have people used more diverse or conformist language over written history? How does the content of a tweet change its likelihood of catching on? And what makes people and regions more likely to discuss and engage in risky behaviors on the internet?
There are limitations and challenges to using language to study cultural change, but these limitations are not insurmountable, and the strategies that I have outlined here can be effective in overcoming them. It is also worth keeping in mind that, if we don’t use language to study human history, what do we have? No other form of records dates back so far or provides such rich information. Language is arguably the best medium we have—or will ever have—to understand changes across human history, so it is worth making sure that we are studying it with the best methods at our disposal.
