Researchers who cannot read the languages of the countries they study must rely on international media or hire native-speaking collectors. Scholars in the Global South often can afford neither, so data collection itself becomes a source of methodological inequality. This article reports an experiment in closing that gap. Using a large language model as an agentic research assistant, I audited a dataset of Russian leaders’ foreign visits (1991–2023) that my co-author and I had built from English-language sources without reading Russian. The model read every issue of the daily Kommersant from 1992 to 2023, checked the dataset against the Kremlin’s official trip archive, and logged each proposed change with a verifiable source, while all coding decisions remained mine. The audit recovered 134 missing visits and removed or corrected 76 records. The gaps were concentrated in the 1990s and among prime ministers, where international coverage was thinnest. Re-estimating the models of the original article with the revised data left its main finding intact: domestic instability raises the probability of a Russian official visit to autocracies but not to democracies. I draw lessons about where English-language media suffice, where they do not, and how LLMs can lower the cost of multilingual data collection for under-resourced scholars.
Researchers who study countries whose languages they do not read face a basic problem in collecting data. They can rely on international media, or they can hire collectors who know the language. The first route requires institutional subscriptions to news databases such as Factiva or LexisNexis. The second requires funding. Scholars in the Global South often have access to neither (Tickner 2013; Medie and Kang 2018; Öztığ 2022). Libraries, particularly at peripheral universities, rarely subscribe to the major news databases (UNESCO and ISSC 2010). Research grants are scarce, and when they are secured, they pay research assistants very little (Steinert et al. 2021). Data collection itself thus becomes a source of methodological inequality in global academia (Pasha and Murphy 2002). My co-author and I ran into these barriers when we built a dataset of Russian leaders’ foreign visits (Balcı and Aras 2025). My co-author solved the staffing problem by recruiting graduate students to collect the data. A second problem remained: none of us read Russian. We assumed that high-level visits are among the most visible acts of a state, so international English-language media would capture them. We ran several reliability checks, but the data were still incomplete.
For full text, click
Bir yanıt yazın