One of the great challenges in autonomously identifying competing narratives in the news lies in the complexity of understanding that two articles are not just discussing the same topic, but are presenting alternative framings and contextualizations around that topic. For example, take yesterday's CNN headline about a breaking study about the efficacy of the Pfizer and Moderna vaccines against the South African Covid-19 variant: "Lab studies suggest Pfizer, Moderna vaccines can protect against coronavirus variant" that opens with "A new report published in the New England Journal of Medicine on Wednesday suggests that Pfizer-BioNTech's Covid-19 vaccine can protect people against concerning new coronavirus variants, including one first seen in South Africa called B.1.351."

Similarly, the WSJ's coverage was headlined "Pfizer-BioNTech Vaccine Is Effective Against South African Variant, New Research Shows" and opened with "The Covid-19 vaccine from Pfizer Inc. and BioNTech SE was shown to generate protection against the variant first identified in South Africa in laboratory testing, according to research published Wednesday on the New England Journal of Medicine website. The results suggest the vaccine generated a slightly lower immune response against the variant's mutations than the original strain circulating in the U.S., but was still effective at neutralizing the variant virus, according to the research. The finding is consistent with earlier research of Pfizer’s vaccine against emerging variants."

Both articles offered fairly optimistic framings that the vaccines were sufficiently effective against the variant. This was shared by a broad swath of coverage.

In contrast, a competing narrative across another swath of coverage offered a more somber framing, such as NBC's syndication of Reuters' "Pfizer says South African variant could significantly reduce vaccine protection: A lab study suggests the South African variant of the coronavirus may reduce antibody protection from the Pfizer Inc/BioNTech SE vaccine by two-thirds." that opened with "A laboratory study suggests that the South African variant of the coronavirus may reduce antibody protection from the Pfizer Inc/BioNTech SE vaccine by two-thirds, and it is not clear if the shot will be effective against the mutation, the companies said on Wednesday." The article went on to note "Because there is no established benchmark yet to determine what level of antibodies are needed to protect against the virus, it is unclear whether that two-thirds reduction will render the vaccine ineffective against the variant spreading around the world." and ended with "Moderna also said the actual efficacy of its vaccine against the South African variant is yet to be determined."

Interestingly, the original Reuters story is titled only "Pfizer says South African variant could significantly reduce protective antibodies," meaning it was NBC's own editorialization to add the subtitle "A lab study suggests the South African variant of the coronavirus may reduce antibody protection from the Pfizer Inc/BioNTech SE vaccine by two-thirds." When The Guardian syndicated the same story, it subtitled it with the more optimistic "Study finds fall in antibody activity – but scientists say jab should still protect against severe disease and death."

While sentiment analysis might be able to help separate some of these narratives, sentiment systems are limited to looking at the emotional connotations of words and lack the domain knowledge to understand that "may reduce antibody protection by two-thirds" is a much more negative statement than its surface words might immediately suggest.

Instead, what is needed are NLP approaches that can take these headlines and articles and understand that they are offering competing framings around the vaccine study, with one narrative framing focusing on a more positive takeaway that the vaccines are sufficiently effective, while the others focus on a more somber framing that efficacy was reduced and in the absence of scientific understanding of minimal efficacy for Covid-19, it is uncertain how well the vaccines protect and if they may only protect against the most severe cases.

A system that could automatically identify such competing framings could allow us to map out the narrative space around pandemic developments to help guide public health officials in understanding the public information environment and clarifying points of contention in the news.