Abstract
This paper investigates how English-centric developments of computational text analysis methods (CTAM) shaped the experiences and practices of political scientists and other social science researchers working with English, non-English, and multilingual texts prior to the wider adoption of large language models (LLMs). Drawing on a survey of 433 scholars who published text-based research in top social science journals between 2016 and 2020, we document the pre-LLMs landscape, presenting scholars’ concerns about CTAM’s validity and availability, use of validation strategies, and barriers to working with multilingual corpora. Scholars working with texts in multiple languages expressed greater concern about CTAM’s validity yet did not report using more validation strategies. Furthermore, non-native English-speaking scholars were more likely to say that they rely on English-language texts when using CTAM than when not. We argue that these patterns reflect a structural bias operating through individual scholars’ choices that aggregates into a systematically distorted field-level research record. Building on these results, we then reflect on whether generative LLMs have altered this landscape and argue that while technical entry barriers have been lowered, the underlying structural inequalities persist, now less visible because they operate inside opaque models and not just explicit methodological choices. In particular, the risk of losing sight of validation is greater, not smaller, in the age of rapidly developing methods.
| Original language | English |
|---|---|
| Journal | Acta Politica |
| DOIs | |
| State | Accepted/In press - 2026 |
Bibliographical note
Publisher Copyright:© Springer Nature Limited 2026.
Keywords
- Computational text analysis
- Multilingualism
- Text-as-data
Fingerprint
Dive into the research topics of 'Challenges for multilingual computational text analysis researchers: evidence from a survey of social scientists'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver