Polarization as a linguistic feature: a systematic review of computational operationalizations of polarization

This systematic review maps associations between linguistic features and computational methods used to operationalize polarization in text data. Using the PRISMA method, the review searched the bibliographic databases Web of Science, Dimensions, and Elicit as of 03.11.2025, yielding 63 publications after screening. Data extraction captured target concepts, linguistic indicators, computational methods, and publication characteristics. The synthesis uses descriptive and relational approaches to examine frequencies, temporal trends, and associations between concepts, indicators, and methods. Findings show distinct patterns in affect-laden indicators, especially sentiment and emotion, which are often combined with pronoun-based group-reference measures. Rule-based approaches are still widely used for grammatical features, while recent work increasingly uses pretrained or large language models for context-dependent indicators. Future research should validate key indicator combinations and relate them to established polarization measures beyond text-as-data research.