The Stanford Computational Antitrust Project announces the publication of “Competition in the Press: A Computational Analysis of U.S.–German Newspaper Discourse (1870–1945)” by Anselm Küsters. The article appears in Volume 6 of Stanford Computational Antitrust (pp. 68–111).

What do people mean when they talk about competition? Küsters answers the question with a corpus of approximately 1.1 million digitized newspaper articles published in the United States and Germany between 1870 and 1945. The analysis proceeds in three steps. Keyword frequencies establish the vocabulary surrounding competition in each national press. Dynamic topic modeling tracks how thematic clusters shift across decades. Cross-lingual semantic embeddings then measure where competition discourse sits between economic-administrative vocabulary on one side and war-and-contest vocabulary on the other.

The two countries diverge sharply. In the German press, competition migrated from civic-commercial vocabulary in the 1870s toward organized performance, order, and martial imagery by the 1930s. In the American press, competition stayed anchored to institutional and economic vocabulary throughout the period. The divergence predates the Sherman Antitrust Act of 1890. Küsters reads this as evidence that the two legal regimes were built on pre-existing semantic foundations rather than the reverse.

The article connects the historical analysis to present-day enforcement. Küsters traces a continuity from the German press’s framing of competition as a rule-bound contest to the legal doctrine of Leistungswettbewerb, the German concept typically translated as competition on the merits.” He then argues that machine-learning screening tools and large language models trained predominantly on one national tradition will reproduce that tradition’s implicit frames. A model trained on U.S. enforcement records learns to associate harmful conduct with terms such as conspiracy, foreclosure, and market share. The same conduct described in German or EU text data may be associated with orderly competition, fair trading, and performance. As enforcement agencies move toward algorithmic screening, this cross-lingual semantic mismatch becomes a concrete risk.

Thibault Schrepel, founder of the Stanford Computational Antitrust Project, stated:

“Küsters shows that the semantic foundations of competition law diverged across jurisdictions before the statutes were written. He then closes the loop with a warning that matters now. Algorithmic enforcement tools inherit those foundations from their training data. Agencies adopting these tools without auditing their jurisdictional coverage will import one tradition’s assumptions about what competition is into another tradition’s enforcement decisions. That is the kind of finding the field needs at this moment.”

Anselm Küsters is affiliated with the Max Planck Institute for Legal History and Legal Theory in Frankfurt am Main and with the Centre for European Policy in Berlin. The article is available for download on the Stanford Computational Antitrust project’s page.