A pilot corpus study of stance and compression in English-language digital news headlines

Abstract

This article examines how a small corpus can still reveal stable editorial preferences in digital news headlines. The study uses a pilot corpus of 42 English-language headlines collected from the Reuters Technology page and the Kun.uz English homepage on April 6-7, 2026. The analysis combines simple corpus counts with close qualitative reading. Four markers were coded: headline length, numerals, attribution verbs, and explicit geographical references. The results show that the two datasets are close in average length, but they package public reality differently. Reuters headlines are strongly corporate and market-centered, while Kun.uz foregrounds state institutions, public regulation, and national framing. Numerals occur more often in the Kun.uz sample, whereas Reuters relies more heavily on company names, deal vocabulary, and high-speed risk framing. Because the corpus is small and opportunistic, the findings are interpreted cautiously. Even so, the pilot design demonstrates that corpus methods help describe news style with evidence rather than intuition.