Art courtesy of Malia Kuo.
The various groups we belong to are at the heart of our human identities. These social identities shape who we are and influence everything from our individual actions to our shared interactions. We behave in accordance with the norms of the groups we identify with, with group memberships that are dynamic. This means that our social identities are context-dependent; at any moment, the group membership that is psychologically salient for a person can change. In other words, our actions and interactions often conform to the norms of the group we identify with at a particular instance.
This prompts questions about our social identities: what factors determine which groups we identify with? How do we switch between our different identities, and how do we deal with competing identities? The major roadblock to studying these concepts is that social identity salience is difficult to assess, especially in a natural setting. Researchers face obstacles and uncertainties in determining which identity is guiding someone’s actions and interactions at a given moment.
In a creative attempt to address this, researchers from the University of Exeter in the United Kingdom recently developed an analytical protocol, ASIA (Automated Social Identity Assessment), that uses linguistic indicators in text to infer salient group membership at a particular moment.
“Salience is a very dynamic thing, so it can switch very quickly,” said Miriam Koschate-Reis, the lead researcher for this project. “If my little girl runs through here, then I can very quickly switch to being a parent and back to being an academic. It’s very fast switching, and we haven’t really looked much at these switches.” (Ironically, seconds after she said this, her little girl did run through the room.)
In the past, researchers have attempted to use self-reporting as a way of assessing salient social identity. However, self-report measures are not very useful for studying temporal dynamics of social identities in natural settings, nor are they reliable for studying long periods of time, because they provide relatively limited datasets. Additionally, if someone is asked to self-report their “main” social identity at a particular moment, they may lack the introspection to actually answer the question. For example, take a moment now to try to identify your salient social identity—student, feminist, daughter, etc. It is quite difficult.
In contrast, ASIA relies on computational linguistics and uses a binary classification model in order to determine which of two social identities is salient in a person at a particular moment. ASIA makes use of linguistic indicators, because sociolinguistic theories assert that both vocabulary and stylistic choices are affected by social variables and groups. Linguistic information has been established as a reliable way to determine group classification and identification. Additionally, there is a wide availability of useful data in the form of written text in online forums, which makes models that focus on linguistic styles convenient and desirable.
However, this data use doesn’t come free of concerns. While developing the ASIA protocol, the researchers made sure to center ethical considerations—mainly the ethical implications of assessing specific social identities in the first place, as well as of using online data to train and validate the model.
“It’s quite tricky when you’re almost looking into people’s minds. …. . . that’s why we really felt strong about writing and putting ethics first to say, if you want to develop this tool, please think carefully about the ethics of it,” Koschate-Reis said.
Koschate-Reis and colleagues concluded that researchers have a responsibility to consider any foreseeable harm to individuals, especially those whose identities may expose them to discrimination and ostracism. They explained that public online forums are perhaps a more ethical source of data than social media platforms where users face difficulty or confusion in selecting appropriate privacy settings. Public online platforms generally have anonymous users with little personal identifying information.
After training and testing their program, the researchers concluded that ASIA provides a mechanism to assess salient social identities using naturally occurring data at a scale large enough to investigate salience within a person over time. This step is essential in investigating how people switch between different identities and when those identities come about. For example, consider: when does someone start to identify themselves as a parent? Is it when they become aware of pregnancy, when they are buying child-raising materials, or when the baby is actually born? The ASIA protocol gives researchers a reliable way to seek answers to these questions, furthering the general effort to learn about our social identities in natural social settings.
Koschate, M., Naserian, E., Dickens, L., Stuart, A., Russo, A., & Levine, M. (2021). ASIA: Automated social identity Assessment using linguistic style. Behavior Research Methods. doi:10.3758/s13428-020-01511-3