Analyzing the Language of AI

By the time you finish reading this story, these words will have affected the way you think about Artificial Intelligence.

The way we communicate about artificial intelligence (AI) affects our opinion of it. But how can we quantify the effect that one written article has? How about 1,000 articles? And here’s the trippy part: Is AI affecting its own discussion? These are perfect questions for Trinity University communication professor Althea Delwiche’s lab, where undergraduate researchers are hard at work analyzing the language—with both positive and negative characterizations—that’s circling this high-tech topic right now.

“This is a cutting-edge lab where computer science and communication intersect. We are analyzing the effects of powerful large language models on the media coverage of AI,” says researcher Matvei Popov ’25, a computer science major from St. Petersburg, Russia. “What is affecting the media coverage of AI? How do people talk about it? What is their sentiment and how does the public opinion shift over time? Are we all doomed?”

Not to leave you hanging—because the jury’s still out on that last one—but the good news is that Popov and his classmates are compiling some impressive research on the other questions, thanks to high-level faculty support and Trinity-style interdisciplinary collaboration.

Popov, and the rest of Delwiche’s team, is primarily focused on a technique called large language model-assisted content analysis. That’s “LACA” for short.

This is quantitative communications research that, in simple terms, counts the number of times events occur—events such as the use of a word or phrase, or the emergence of larger patterns of language. Before the rise of AI, Delwiche, Ph.D., says, quantitative content analysis was a “useful but grueling” task for researchers, and, to boot, “it does not scale well in a world of large data sets.” But now, current research relies on large language models (LLMs), which are artificial neural networks (a type of AI) that can process and classify language.

But now, researchers like Popov, alongside teammates including Mylo Mittman ’27, a communication major from Dallas, and Gayatri Rajamony ’25, a computer science major from Austin on the pre-law track, have access to a new world of technological tools that can rapidly accelerate the pace and scope of this type of research.

This semester, Popov and the rest of Delwiche’s team are busy compiling a huge number of news articles that mention AI—a part of the project that heavily relies on the communication students in the lab. The next step for the team is developing a tool called a content analysis codebook, which is a refined framework for sifting through and categorizing these mentions of AI as either negative or positive.

Ultimately, the group will use various LLMs in conjunction with this codebook (and their dataset of articles) to find patterns in the language surrounding AI. Each week, the group huddles into a snug conference room deep in Trinity’s Richardson Communications Center, discussing research goals and assigning tasks.

The lab relies on teamwork: its computer science majors share their growing coding and programming knowledge with the communication students, who in turn lend their burgeoning expertise in media analysis and language. 

“Trying to apply computer science to communication is very challenging but also rewarding.

If I’m trying to train a machine learning model to understand communication without understanding it myself, it is likely not going to end well,” Popov explains. “Without the comm majors there, it would take me a year and a half, probably, to understand some of these concepts.”

Being part of the lab is a distinctly tech-focused experience, but it’s one that revolves around human connections across disciplines, says Mittman, one of those communication majors who’s helping the team with data collection and cleanup.

“Having the opportunity to look at things from multiple perspectives makes me excited about learning,” Mittman says. 

For Rajamony, who is planning to pursue a career in patent or copyright law, this interdisciplinary research is the exact type of hands-on opportunity that made coming to a liberal arts school like Trinity so appealing in the first place.

“I knew pretty much immediately that I wanted a smaller liberal arts college, a place that valued the interdisciplinary nature of my interests. And it’s just always nice to see other people’s perspectives on computer science stuff, such as having a communication perspective,” Rajamony says. “It’s really useful to me as a computer science major who wants to make ethical decisions.”

One of the biggest links for these students, across all their different disciplines and interests, has been the chance to work with a driven, dedicated faculty member like Delwiche.

Mittman, who found the lab after taking a previous Delwiche course on AI, loves the professor’s openness to collaboration, stating, “She’s very accessible, and when it comes to the lab, she’s willing to talk to everybody and hear everybody out. She’ll say, ‘If you have an idea, just let me know, and we’ll organize it with the rest of the lab to see how that can fit into our bigger project.’”

On the horizon for the group? Continuing to experiment with (and train) a series of LLMs that can perform the operations the group needs to start analyzing all the data they’re collecting.

“We’re basically training (these LLMs) to perform text analysis, which can help us identify the spectrum of sentiment about AI over time,” Popov says. “The better we can fine-tune and customize these models, the closer we will get to our desired analysis; that’s definitely been kind of a challenge, but I think we are very close to solving it.”

Rajamony, who spends her days working on automating the process that “feeds” these models in the hopes of speeding it up, says the challenge remains to figure out how the group can take this model and adjust it to perform more with what we value. 

Behaving, training, feeding—if we’re examining the group’s language around AI, Mittman says, you might notice a pattern developing already.

“My initial impression of AI is that yes, I think it’s cool, but it’s still kind of like a baby,” Mittman explains. “We are teaching it, we are tweaking it, and we are basically helping the baby learn, for lack of a better term. That’s what I think is really cool about this project, too: that we’ll be able to contribute something meaningful to... the growth of this field.”

And as for the fields of computer science and communication, this project is another perfect example of how growth at Trinity happens best when it happens together.

“I’m so glad that this lab has so many people of different backgrounds and experiences,” Popov says, “because everybody can address different problems and create some really cool ideas by working together. When different majors can combine together on solving big problems, that’s just fascinating.”

For Mittman, this is the type of research that is more than a line on a résumé.

“I’m in my third semester, and it’s honestly amazing to me that if things go right in this lab, which I bet they will, I could possibly be part of a team that publishes a paper—a project that I helped take these next steps in research for such a developing field, such a new field, such a field that some people are excited about but scared of,” Mittman says. “I think one of the coolest things about what we’re doing is that hopefully, in some small way, we’ll have a hand in shaping public perception of this emerging technology.”