Modeling legitimation in discourse: a corpus-based textual analysis of government news

Legitimation strategies are communication efforts aimed at securing public or stakeholder approval, yet large-scale analysis of these strategies is limited. This study uses Latent Dirichlet Allocation (LDA) topic modeling to identify prominent topics as the potential ‘salient elements’ of a discourse that are legitimized across two comparable corpora of government COVID-19 news from Hong Kong and Mainland China. The corpus comprises 4,143,250 words, including 2,304 Mainland Chinese news articles (1,034,698 words) and 5,662 Hong Kong Chinese news articles (3,108,552 words). The LDA results reveal significant differences in the topics and legitimation strategies between the two regions. The Mainland China government employs more focused topics to establish legitimacy, mainly leveraging ‘authorization’ and ‘moralization’ through the dominant theme of ‘prevention and control,’ whereas the Hong Kong government relies on ‘rationalization’ and emphasizes ‘detection’ initiatives. Additionally, the implementation of the same legitimation strategy varies, as reflected in different ways of government self-presentation, such as being achievement-focused versus action-oriented. These findings provide linguistic evidence of how governments in a shared national context adopt different legitimation strategies, influenced by unique administrative characteristics. This study demonstrates the use of automatic textual analysis as a complementary method for traditional discourse analysis and highlights its effectiveness in large-scale, corpus-based research. It presents a replicable approach for analyzing discourse-level strategies, with practical implications for public health communication and crisis management.