🇺🇸 Emily Ocasio

The question

Newspapers do not report every murder the same way. Some victims are described with the details that make a person legible — a job, a family, a nickname, a line from a grieving friend. Others are a name, an age and an address.

Ocasio set out to measure that difference rather than assert it, using artificial intelligence to analyse humanising language across the Boston Globe's homicide coverage from 1976 to 1984, and reporting racial disparities in how victims were written about.

The distinction has consequences beyond the page. How a killing is described shapes which cases attract sustained coverage, which draw political attention and which are quietly closed, so the language of a homicide report is not a stylistic matter but part of how a city allocates its concern.

Why the method matters

Claims about media bias are usually argued from examples, and examples can be selected. A computational approach changes the terms: it processes the whole archive, applies the same criteria to every story, and produces a distribution rather than an anecdote.

The choice of period is deliberate too. Coverage from 1976 to 1984 is old enough to be a closed record and recent enough to belong to the modern newsroom. The press release does not publish her corpus size, model, coding scheme or effect sizes, so the strength of the finding cannot be assessed from this record — only the finding as the organisers reported it.

Second in the oldest competition in America

The Regeneron Science Talent Search was founded in 1942 and 2023 was its 82nd year. Nearly 2,000 students entered, 40 became finalists, and $1.8 million was shared among them; $3.1 million was distributed in total including school awards. The gala was livestreamed, hosted by the broadcaster Soledad O'Brien.

First place and $250,000 went to Neel Moudgal of Saline, Michigan, for a model predicting RNA structure. Third and $150,000 went to Ellen Xu of San Diego for a smartphone-based algorithm to help diagnose Kawasaki disease.

Humanities questions, scientific tools

A project about newspaper language winning $175,000 in a science competition says something about where the boundaries now sit. The competition rewards research method rather than subject area, and a well-built computational analysis of a text archive is research by that standard.

It also belongs to a wider shift. Digitised archives have made it possible to test questions about representation, framing and attention that were previously matters of argument, and the tools are the same ones used on genomes and telescope images. The discipline has a name — computational social science — and it is unusual mainly in how rarely it appears at high-school level.

The judging format at the Science Talent Search rewards that kind of crossing. Finalists spend a week in Washington being questioned by scientists outside their own field, on the reasoning behind the work rather than on its subject, and a project has to survive interrogation from people who do not share its assumptions.

What is documented — and what is not

Firm: second place, the $175,000, her age, her home town, the newspaper, the period studied and the reported conclusion, all published by the Society for Science in March 2023.

Open: the technical detail behind the analysis, any peer-reviewed publication, and what Ocasio has done since. None of it appears in the sources gathered for this page.

For a reader searching her name, the honest summary is that one thing is firmly established — a major national award for a rigorous piece of computational media analysis — and that everything downstream of it, from publication to career, is simply not in the material available here.