New ideas in biomedical science are traveling less far than they used to. Concepts introduced in 2009 spread roughly half a standard deviation less broadly beyond their home field than concepts introduced in 1980, and a growing reliance on jargon appears to be a big part of the reason.
The study comes from economists Enrico Berkes of the University of Maryland, Baltimore County and Ruben Gaetani of the University of Toronto. Titled “The Decline in the Transmission of Scientific Ideas,” it is dated July 2026 and was posted as an NBER conference paper.
Tracking concepts across 31 million papers
The authors start from history. X-ray crystallography came from physics and helped reveal the structure of DNA. CRISPR began as basic research on bacterial immune systems and became a gene-editing tool across biology, medicine and agriculture. The biggest payoffs often come when ideas cross fields.
To measure that crossing, they used the full PubMed corpus of more than 31 million papers. They identified new scientific concepts introduced between 1980 and 2009 and followed the papers that adopted each one over the next ten years. Breadth of diffusion was measured as the semantic distance between the abstracts that introduced a concept and the abstracts that later used it.
Language accessibility was measured as the share of words in the introducing abstracts that were common in the literature of the previous five years. Knowledge depth was proxied by how specialized the introducing authors were.
Plainer language travels further
Over the three decades, the share of common words in abstracts introducing new concepts fell by roughly 4 percentage points, and the typical score dropped from 83.4 percent in 1980 to 1994 to 79.3 percent in 1995 to 2009. At the same time, the median distance between introducing and adopting papers fell by about 0.06, around half the measure’s standard deviation.
This was not because science became more uniform. Placebo simulations suggested, if anything, that articles have grown slightly more dispersed over the period.
In the main sample of about 42,000 concepts, a one standard deviation rise in accessibility went with a breadth of adoption 19 percent of a standard deviation wider, before field controls trimmed that by about a third. At the level of individual fields, a one standard deviation rise in field-specific accessibility raised the chance that a field adopted a concept at all by more than 5 percentage points, and the number of adopting papers by more than 10 percent.
Deeper concepts, built on more specialized knowledge, were introduced in harder language and diffused less broadly, with changes in language accounting for a substantial share of that link.
A jargon example and a policy angle
One example is a 1996 Cell paper on the Toll signaling pathway in fruit flies, part of the foundation for Jules Hoffmann’s 2011 Nobel Prize. Its abstract scored 77.9 percent on accessibility. A version rewritten in plainer language by a large language model, keeping the substance, scored 92.7 percent.
The authors build a model in which scientists trade off precise language for close peers against accessible language for distant fields. As knowledge piles up, specialized language becomes the better private choice, and diffusion narrows. Because researchers value recognition from nearby fields more than society does, the model says incentives that reward reaching distant audiences would widen adoption and raise long-run growth.
Study Details:
- Title: The Decline in the Transmission of Scientific Ideas
- Authors: Enrico Berkes, Ruben Gaetani
- Journal: NBER conference paper
- Publication Date: July 2026
- Link: https://conference.nber.org/conf_papers/f245515.pdf
