War Metaphors in Scientific Papers: What 21.4 Million Abstracts Reveal About Why Scientists Increasingly "Go to War"
A Curious Linguistic Phenomenon
Open the abstract of nearly any AI paper and you will likely read sentences like:
- "We attack the problem..."
- "Our method defeats the baseline..."
- "This finding conquers a long-standing challenge..."
- OpenAlex (all disciplines): share of abstracts containing military terms rose 48% between 2010 and 2025
- PubMed (biomedicine): up 32% over the same period
- The two databases' growth rates are highly correlated: r = 0.96, p < 10⁻⁸ — a genuine cross-disciplinary trend, not a database artifact
- Growth accelerated sharply after 2019; COVID and the post-2022 large language model era pushed the trend further upward
- By discipline: social sciences show the highest *levels* of militaristic language; engineering and computer science show the *fastest* growth
- By geography: authors from the Global South show the fastest growth in militarized language
- By language background: the "militarization gap" between non-native and native English speakers narrowed during the COVID and LLM eras—non-native authors "caught up"
- Country-level conflict intensity correlates with militarized language among that country's authors: r = 0.77-0.84
- At the yearly level, global conflict intensity correlates with militarized language in abstracts
- 801 participants
- Each read multiple pairs of abstracts (32,040 trials total)
- Each pair had identical content except one version used war framing ("attack the problem") and one used neutral framing ("address the problem")
- Participants rated credibility, willingness to fund, policy support, and urgency
- Scale plus causality: 21.4 million abstracts establish *what* is happening; the 801-person experiment establishes *why it matters*. This "large-scale observation + small-scale causal" combination is a model for science communication research.
- Cross-database validation: The r = 0.96 consistency between OpenAlex and PubMed rules out single-database bias.
- Real-world linkage: Tying scientific language to UCDP conflict data elevates the work from linguistics to social psychology—scientific language is not an autonomous system; it fluctuates in sync with real-world violence.
- Astonishing scale: 21.4 million abstracts, 16 years, two databases
- Well-designed causal experiment: within-subject design controls individual differences; 32,040 trials give ample statistical power
- Interdisciplinary collaboration spanning bioengineering, communication, and network science
- Small but robust effects: d_z = -0.28 is small-to-medium in psychology, but p < 10⁻²⁰ signals robustness
- The definition of "militaristic language" may be too broad: words like "attack" and "target" have long had neutral scientific uses ("target protein"), so the effect may be overestimated
- The 801 participants may not represent typical readers—are they scientists or the general public?
- The study does not distinguish deliberate from unconscious militarization, which matters for interventions
- The experiment measured only four outcomes; war framing might have positive effects on unmeasured dimensions like memorability or virality
Most readers never notice anything odd. But an interdisciplinary team at the University of Pennsylvania analyzed 21.4 million scientific abstracts (2010-2025, from OpenAlex and PubMed) and found that the use of "war vocabulary" in scientific writing rose 48% (OpenAlex) and 32% (PubMed) over the past 15 years, with growth accelerating sharply after 2019.
More surprising still: using war framing not only fails to make papers more persuasive, it actually lowers readers' trust in the research.
This is not a minor linguistic curiosity—it is a mirror reflecting the psychological state of the entire scientific community.
What the Data Show
Scale and Growth
Who Is "Fighting"
Correlation with Real-World War
The most striking finding: militaristic language in abstracts correlates with real-world conflict intensity. Comparing against Uppsala Conflict Data Program (UCDP) data:
This is not merely "scientists using war metaphors." It is real-world war seeping into scientists' language.
The Causal Experiment: War Framing Reduces Credibility
The team also ran a within-subject experiment:
| Metric | Effect size | p-value | |--------|-------------|---------| | Credibility | -0.18 Likert units (d_z = -0.28) | < 10⁻²⁰ | | Willingness to fund | d_z = -0.12 | significant | | Policy support | d_z = -0.08 | significant | | Urgency | trend-level increase (d_z) | trend-significant |
War framing made readers judge research as less credible, less worth funding, and less deserving of policy support. The only small "benefit" was a trend-level increase in perceived urgency—at the cost of credibility.
It is like writing "I defeated all my competitors on the battlefield" on a résumé: you don't come across as impressive; you come across as boastful or aggressive.
Why Scientists Increasingly "Go to War"
The paper does not offer a single cause, but the data point to several threads:
1. COVID's linguistic shock: The post-2019 acceleration coincides with "fighting the pandemic" becoming mainstream discourse. When "frontline workers" and "the war against COVID" become everyday phrases, scientists borrow that vocabulary unconsciously. 2. The LLM era's competitive narrative: After 2022, the "arms race," "beating baselines," and "attacking benchmarks" became standard AI writing. Computer science and engineering being the fastest-growing fields matches the LLM timeline. 3. Linguistic convergence by Global South authors: Non-native English authors often learn academic writing by imitating high-frequency papers—which are precisely the militarized AI and biomedical papers. If your model texts say "attack/defeat," you will too. 4. Semantic seepage from real war: The r = 0.77-0.84 correlation with conflict intensity suggests scientists do not write in a vacuum; their news, social feeds, and conversations are saturated with war discourse. Notably, experiment participants attributed their judgments to "the evidence itself," not the framing—an unconscious process.
Methodological Strengths
Honest Assessment
Strengths
Limitations
Implications for Scientific Writing
The paper's direct advice to scientists: use less war framing. Not for reasons of political correctness, but because it genuinely lowers the probability your paper will be trusted. When you write "we attack the problem," readers—including reviewers and funders—subconsciously judge your research as less credible.
The deeper question: the way we write papers is shaped by the era we live in. The COVID era, the LLM era, the era of geopolitical conflict—these are not neutral backdrops. They seep into our vocabulary, our thinking, our scientific judgment.
Scientists often claim objectivity and neutrality, but this study, using 21.4 million abstracts, shows that no scientist is truly neutral—we are all products of our times, down to our abstracts.
---
Paper: https://arxiv.org/abs/2606.23462 HTML version: https://arxiv.org/html/2606.23462v1 Authors: Sovesh Mohapatra, David Lydon-Staley, Dani S. Bassett (University of Pennsylvania) Code: No official code repository provided