Natural language processing can be used to look at annual reports and ask: What exactly does the document say? How is the information presented? (Getty Images/shapecharge)
Corporations are increasingly disclosing their diversity and disability policies in their annual reports. While the language may seem meaningful, it’s still highly biased, says Jacqui Gagnon, assistant professor and CPA research scholar at the University of Regina’s Hill School of Business and Levene Graduate School of Business.
Using natural language processing, the CPA explores repetitive use of seemingly innocuous language patterns to uncover trends that contribute to unconscious bias and the perpetuation of stereotypes.
She will share the results of her research at CPA Canada’s ESG Symposium.
CPA Canada: Why are you interested in natural language processing and how does it work?
Jacqui Gagnon (JG): My motivations are twofold: personal, because I have a black child and a child with a disability, so I feel directly connected, but also professional, because I see organizations struggling with socially sensitive disclosures and want to help them address these issues. For example, being more sensitive to people with disabilities.
My research uses natural language processing, which combines linguistics, computer science and artificial intelligence. In practice, it involves selecting large samples of data and using computers to analyze words in context—in my case, the 150 words that precede and follow.
This gives us a better idea of how companies frame issues: what they talk about, the language they use, and what that language really means, which is impossible to do one document at a time. Through repetitive use in different places, these language patterns change how we perceive certain issues.
CPA Canada: Can you give us examples of disclosures that you feel are biased?
JG: Let’s imagine a job posting that says “applicants will be given equal consideration based on merit, regardless of race, gender, marital status, etc.” This may seem like a straightforward sentence, but the way it is phrased links merit to diversity and suggests that those with diverse characteristics may not have a chance.
Furthermore, what is merit? Is it education? Training? Experience? Only those in positions of power know.
Another point is the use of “regardless of” phrasing, as if institutions were neutral, when in fact they suffer from structural and unconscious human bias in their decision making. Let’s face it—it is very difficult to be impartial.
Disability commentary often reproduces legislative language, adding unnecessary clutter that contributes to length without adding substance.
CPA Canada: Why is disclosure on diversity and disability so important today?
JG: it’s important because there is clearly room for improvement. For many organizations, disclosure is still just about appearances. To look good, they use boilerplate language in their reports, as if they were checking off boxes. This is what we call social washing.
At the same time, there are more laws on diversity and disability disclosure now than ever before. Social responsibility issues are critical because they impact not only employees, but society as a whole. So addressing corporate language is a great place to start.
CPA Canada: How do you recommend making a difference?
JG: No one expects organizations to get it right the first time, but they need to approach the issue honestly.
Rather than saying, “We don’t do this or that,” organizations need to admit: “We have these biases and here’s how we’re addressing them.” How? By quantifying things. For example: How many racially or ethnically diverse employees work in the organization? How many people with disabilities? In what types of positions? And so on.
We also need to combine the data. The number of workers with disabilities by gender, for example, is very important for honest reporting. That’s how you can make a difference: by knowing where you’re starting from and where you want to go, so you can better target the steps you need to take to get there.
CPA Canada: What other uses are there for natural language processing?
JG: Natural language processing can reveal a lot of information about corporate social responsibility, such as compensation, for example. You can use natural language processing to look at annual reports and ask: What exactly does the document say? How is the information presented? It also allows you to cross-reference what organizations say with what the people who work there think, whether they are able-bodied or have a disability.
Many people in our society do not have the skills to read financial reports, but natural language processing can identify areas of concern and help companies take steps to make reports more readable for the general public. Annual reports link audited financial information with management discussion, creating an ideal document for honest and intentional communication. But too often management commentary is long and dense. Essentially, the more we can provide clear information to the public, the more accountable we will be.
Before becoming a teacher-researcher, I worked in public practice. I know there are many challenges, but I also see the opportunity for organizations to take a big step forward. There has never been a better time.
To learn about where Canadian public companies stand on social reporting, see State of Play: Study of social disclosures by Canadian public companies and The “S” in ESG: an increasing focus for organizations. Plus, find out more about the business impact of environmental and social issues, how CPAs can lead ESG initiatives, and how SMPs can build an ESG strategy.