A new paper by Tessa Charlesworth, assistant professor of management and organizations at the Kellogg School, explores this complex question by computationally analyzing language use over 115 years, from 1900 to 2015. Specifically, Charlesworth and her colleagues — Nishanth Sanjeev of New York University and Mark L. Hatzenbuehler and Mahzarin R. Banaji of Harvard University—examined how the characteristics associated with various social groups have changed over time.
They explored both the manifest and implicit meanings of these stereotypes. “I often give the analogy of an iceberg,” says Charlesworth. “There is the tip of the iceberg that we can see above the water line. These are the actual words we use to describe different groups,” what the researchers call “manifest” content. “But hidden beneath the surface of the water are the hidden meanings, such as how positive or negative, or capable or incapable these words are”—the “latent” content.
Overall, the researchers found, group stereotypes have changed significantly in their manifest content, but their latent content has remained much more stable. For example, “you can think of some archetypal examples of how our stereotypes of Black Americans have changed over time, from lazy in the 1900s to helpless in the 1990s,” explains Charlesworth. “It’s a different word, but it has the same meaning of helplessness and negativity. We can think of similar examples with women — it was once said hysterical; Now is Emotional.”
For Charlesworth, this pattern suggests that surface-level descriptors may change in meaningful ways, but deep-seated feelings and beliefs are more stubborn.
“It’s a really interesting social phenomenon,” he says. “Society can reinvent itself and, on the surface, pretend to be changing and making progress—despite the fact that there are hidden messages that continue to persist.”
How word embeddings capture social biases
To understand how stereotypes have and haven’t changed over time, the researchers used word embeddings. This type of computational text analysis involves representing words in space based on how often they occur together in a given body of text. Word embeddings allow scholars to measure the relationship of two words based on how close or far they are to each other. The technique allows a computer system that has no idea what the words mean to determine it dog is closer to Cat than is to refrigerator; It’s also how systems like ChatGPT learn to generate such human-sounding text.
Previous research has shown that word embeddings also correlate with experimentally documented biases. For example, the Implicit Correlation Testing shows that people more quickly associate youth-related terms with pleasantness than with unpleasantness (and vice versa for age-related terms). Word embeddings show the same patterns: youth-related terms are closer to words like pleasant rather than words like unpleasantwhile the opposite is true for conditions related to older age.
And because word embeddings can be quantified numerically, changes over time can also be quantified numerically—which is exactly what Charlesworth and her colleagues did.
By examining the characteristics most closely associated with different social groups over time, as well as the latent meaning of those characteristics, researchers could understand how stereotypes have and have not changed.
A century of linguistic change
To begin their study, Charlesworth and her colleagues assembled a vast corpus of texts that spanned 115 years and included works of fiction and nonfiction. Their dataset included Google Books, The New York Times file and Common Crawl, a huge repository that has been pulled from the internet.
They then came up with a list of social groups they wanted to study. The 72 groups they selected fell into four different categories: sociodemographic groups (e.g., black, white, young, older, gay, straight), body-related (e.g., fat, thin, disabled, handicapped), body-related with mental health (eg, depressed, happy, bipolar) and occupational (eg, employed, unemployed, educated, uneducated).
The researchers also created a list of synonyms for each group, making sure to include a range of historically specific terms. For example, the term schizophrenic it wasn’t popular until the early 19th century, so they were included as well psychosiswhich was commonly used to describe the same set of symptoms before that time.
They then compared how these 72 social groups (and their synonyms) related to about 600 trait adjectives that have been widely used in other psychology research.
For each decade between 1900 and 2015, they determined which ten trait words were closest to each of the 72 groups. “They are like words lazy and helplessor are words like hot and kind?” Charlesworth explained.
Finally, she and her colleagues assigned scores—numerical measures of positivity, warmth, and competence—to the top 10 trait words associated with each group at each point in time. This allowed them to calculate how much the latent meaning of the stereotypes changed, in addition to the actual turnover of single obvious words.
Stereotype continuity and change
In all 72 groups, overt stereotypes—that is, the actual language used—changed substantially, while latent content—the underlying associations—remained much more stable. However, notes Charlesworth, “it is not the case that every group changes in manifest content and does not change in latent valence. In fact there is a lot of variability.”
For example, stereotypes of sociodemographic groups changed much more than stereotypes of body-related groups.
Charlesworth and her colleagues suspect that this may have to do with how consistent the stereotypes are at any given time. For example, if the novels, the New Yorkerand National Review everyone portrays the same group in the same way at the same time, there is little room for stereotypes to change in the future. But if there is greater variability between sources, it suggests that the consensus view is breaking down.
And that’s essentially what the researchers found for the body-related and sociodemographic groups, Charlesworth explains: “Everyone says the same negativity about being fat or disabled. And that kind of consistency means that these stereotypes can go unchallenged.” Meanwhile, he adds, “there’s a little more variation in how we talk about sociodemographic groups, which can open the door to social change.”
How often a group was mentioned at all also emerged as a predictor of overt (and to a lesser extent, latent) stereotype change. “The groups we talk about the most will also be the groups that change the most, because there are simply more opportunities to intervene in behaviors we talk about than in attitudes we ignore,” says Charlesworth. “And maybe that’s our best resource for intervention in the future. Groups we can bring to mind in activism and politics will be the ones we can move the needle on.”
Changing hearts (and words)
Charlesworth says the research sheds light on one of the most persistent paradoxes of modern life: despite huge progress for many marginalized social groups, deep inequalities and prejudices persist.
“It resolves some of the ambiguity about how we can both have evidence of some change. . . and really persistent discrimination,” he says. “The words we use to describe these groups change, but beneath the surface, there are hierarchies that are so persistent.”
Figuring out what can be done is an issue he hopes to address in future research. “How do you deal with the latent, underlying meaning of our group stereotypes? How do you disrupt the idea that we feel functional or even legitimate reasons to stigmatize groups?’ she says. “That, I think, will be the key sticking point and the main open question.”
