Flooding Narratives in Pakistan (2025): A Latent Dirichlet Allocation (LDA)-Driven Ecolinguistic Analysis of Dawn News Coverage during the Peak of the Crisis

Abstract

This study examined Dawn News coverage of the 2025 Pakistan floods through a computational ecolinguistic approach that integrated Latent Dirichlet Allocation (LDA) and Computational Grounded Theory (CGT) proposed by Laura K. Nelson (2020). A corpus of 150 e-reports published between during the peak of the crisis (May-September 2025) was processed in Python 3.12 to identify latent themes and framing structures in the media discourse. Standard natural language processing (NLP) and text preprocessing techniques were applied, including tokenization, lemmatization, stopword removal, lowercasing, and vectorization using CountVectorizer method. The LDA model was trained and implemented with five topics specified. The computational results were then interpreted through Stibbe’s (2015) concepts of framing, erasure, and ecological wellbeing.

The LDA analysis revealed five dominant narratives: human impact, infrastructure and environment, relief and housing response, rescue operations, and disaster management. These narratives foregrounded human suffering, state-led responses, and infrastructure fragility, while ecological factors such as deforestation, poor urban planning, and climate mismanagement were erased or silenced in the background. The coverage adopted an anthropocentric framing that represented nature as destructive and humans as passive victims or active rescuers. The study demonstrated how the blending of computational grounded approach and qualitative ecolinguistic interpretation enhance understanding of ecological issues in the modern world. It also underscored the importance of embedding ecological accountability in media narratives to promote sustainability, resilience, and ecological wellbeing.