A few years ago, some of Maria Antoniak’s friends gave birth and recounted the experiences online – in a blog, on Facebook and in email.

“That was my entry into this world of online birth stories,” said Antoniak, a doctoral student in information science, “and I found there were thousands upon thousands out there. I was fascinated. They’re full of emotion, sometimes trauma; sometimes very beautiful, sometimes heartbreaking.”

Antoniak approached this trove of information as a data scientist, performing a computational analysis on nearly 3,000 online birth stories publicly posted on Reddit.

The resulting study is described in “Narrative Paths and Negotiation of Power in Birth Stories,” which will be presented at the Association for Computing Machinery’s Conference on Computer-Supported Cooperative Work and Social Computing, Nov. 9-13 in Austin, Texas. The report sheds light on new mothers’ feelings of powerlessness in the delivery room and illustrates how artificial intelligence tools might be used to analyze relatively complex narratives.

“There’s a common narrative arc across all of these stories,” Antoniak said, “and yet each story is deeply unique. So on the computational side we wondered: Can we detect those narrative arcs? And can we detect the stories that deviate from them?”

Antoniak is first author of the paper, which is co-authored by Karen Levy, assistant professor of information science, and David Mimno, associate professor of information science.

Understanding any but the simplest narratives is a challenge for AI. Birth stories are ideal for training these algorithms because they’re so predictable – the first signs of labor are usually followed by a trip to the hospital or birthing center; a period of intense pain precedes an epidural, or the birth. They involve a standard cast of characters: mother, partner, doctor or midwife, nurses, doulas and family members.

As Antoniak predicted, these similarities made it easier for an unsupervised machine learning algorithm – that is, a model expected to learn for itself what it’s looking for – to generate a list of chronological topics. The list it created began with pairs of words including “weeks pregnancy,” “due induction” and “water broke,” and ended with words such as “cord crying,” “pads clothes” and “breastfeeding day.”

The word “contractions” emerges around a third of the way through the list of topics. “Which makes sense and is so easy to interpret,” Antoniak said. “Usually you would not find such clear patterns.”

The model accurately identified the birth stories that strayed from the usual narrative arc, finding that these stories tended to use phrases such as “emergency c” and “slightly traumatic.” Many less-common birth stories used the phrase “happy ending” – possibly indicating the author’s desire to reframe her story in a positive way.

The researchers also assessed the stories’ power dynamics using a dataset of more than 2,000 verbs, labeled according to whether the verb’s subject is more, less or equally powerful than the object. They found that new mothers tended to view themselves as the least powerful people in the room, other than their newborn babies.

“We wanted to know, from the author’s perspective, who had the power, because a lot of these stories are about decision-making,” Antoniak said. “That might point to relatively small actions that can be taken to make them feel more empowered.”

For example, the results could be shared with medical practitioners, to encourage them to show more empathy to the women giving birth. New mothers could be prompted to tell their birth stories immediately, which has been shown to help restore a feeling of power and reduce postpartum depression, Antoniak said.

“I was very shocked reading these stories because some of them are really very dark and very personal, and women are opening themselves up to share this,” Antoniak said. “I think part of their intention is to help other women, so I thought, anything I can do to highlight these authors’ voices and bring attention to them could be helpful.”

A surprising result of the analysis , she said, was how powerful women considered doulas – trained, nonmedical professionals who support women through the birthing process and assist in around 2% of U.S. births. According to the study, the authors viewed doulas as more powerful than their partners, doctors, nurses and family members – particularly in negative scenarios.

Because the study focused on a single online community, the authors said they were not able to control for demographic variables.

The online birth stories offer a wealth of medical information, which can be difficult for researchers to access because of privacy laws. Patterns revealed by the analysis could potentially inform medical practices. But as far as Antoniak could tell, this study is the first of its kind.

“Quite a bit of past research has been done on the oral telling of birth stories, which of course have existed for millennia, probably as long as humans have been talking,” she said. “But I haven’t found any paper that did a computational analysis of birth narratives, which is surprising because there are so many of them online. And that raises the question: Why aren’t people paying attention?”

The research was supported by the National Science Foundation and the Alfred P. Sloan Foundation.

This story by Melanie Lefkowitz originally appeared at the Cornell Chronicle on November 4, 2019.