Effective leadership communication is crucial during crises for mobilizing public response. The COVID-19 pandemic presented a significant challenge, requiring clear communication of public health guidelines to promote compliance and mitigate the spread of the virus. Prior research indicates that the manner in which instructions are presented significantly affects public behavior, and specific word choice can shape public opinion. Leaders' communication styles are influenced by various factors, including political affiliation, demographics, personal style, and emotional state. The study leverages computer-based corpus linguistic methods to analyze a large dataset of speeches, providing a systematic quantitative approach to identifying patterns and underlying phenomena. Previous research using corpus linguistics has studied language evolution, historical epidemiology, and political trends. This study aimed to explore the associations between COVID-19 case rates and the linguistic properties of governors' speeches across space and time, investigating how pandemic intensity influenced both the content and style of communication, including linguistic markers of stress.

Existing research demonstrates the significant impact of leadership communication on public compliance with health guidelines. Studies show that the way instructions are presented influences behavior, and word choice is strategically used to shape public opinion. Prior work has linked leaders' word choices to political affiliations, demographics, personal styles, and emotional states. Corpus linguistic methods have been successfully used to analyze large text corpora, revealing patterns in language evolution, historical epidemiology, and political trends. However, research specifically examining the relationship between crisis intensity and the linguistic properties of leaders' communication, especially regarding stress markers, is limited. This study bridges this gap.

The researchers assembled a corpus of 1515 speeches (4,049,146 words) from all 50 US state governors, collected from February 27, 2020, to July 14, 2020. Data sources included governors' offices, commercial transcription services, YouTube, and Facebook. Transcripts were curated to include only the governors' words. COVID-19 case rate data per 100,000 persons were obtained from the New York Times. Words used by 20 or more governors, representing at least 0.02% of all words in at least one state, were included in the analysis. Two independent raters grouped words into semantic categories, excluding words with multiple meanings or those difficult to categorize. Words were also grouped by part of speech using the Natural Language Toolkit. Linguistic complexity measures (average word length and syllable count) were calculated. Associations between linguistic features (semantic categories, parts of speech, word length, and syllable count) and COVID-19 case rates were analyzed spatially (across states) and temporally (across weeks within states). Spatial analyses used normalized word counts (percentage of all words spoken by a governor in a state). Temporal analyses used normalized word counts (percentage of words spoken in a given week). Spatial analyses calculated confidence intervals around Spearman's Rho values. Temporal analyses used the standard formula for a distribution of Spearman's Rho values. Polynomial curves were fit to temporal plots for visualization. Only states with at least 50,000 words of speeches were included in temporal analysis to ensure sufficient sample size.

The analysis revealed significant spatiotemporal associations between COVID-19 case rates and various linguistic features of the governors' speeches. Positive associations were found between case rates and: hospital-related words (e.g., "ICU," "ventilators"), negation words (e.g., "can't," "no"), strict instructions (e.g., "prohibited," "compliance"), negatively descriptive words (e.g., "terrible," "worst"), religious words (e.g., "pray," "God"), and extreme descriptive adjectives (e.g., "dramatically," "extraordinarily"). Negative associations were observed between case rates and: job-related words, travel-related words, formal communication words (e.g., "announcement," "declaration"), helpful action words, emergency-related words, and preventative measure words. Regarding parts of speech, positive associations were found with past-tense verbs, present-tense verbs, and adverbs. Negative associations were found with possessive pronouns, plural nouns, and base-form verbs. A strong negative association was observed between case rates and average word length and syllable count, particularly in states with the highest case rates. This suggests a shift towards simpler language as the crisis intensified.

The findings indicate a clear link between the severity of the COVID-19 pandemic and the linguistic characteristics of governors' communication. The increase in hospital-related terminology, negation, and strong adjectives aligns with the escalating crisis. The decrease in job-related and travel-related words reflects the changing context of the pandemic. The unexpected negative correlation with emergency-related and preventative measure words may be due to a shift in communication focus, rather than a decrease in the importance of these topics. The shift from future-tense to past- and present-tense verbs reflects a move from planning to action. The simplification of language (shorter words, fewer syllables) is consistent with a stress response, potentially impacting communication effectiveness. These findings provide insights into how leaders adapt their communication under pressure, highlighting potential areas for improving crisis communication strategies.

This study demonstrates a strong association between COVID-19 case rates and the linguistic features of governors' speeches, both spatially and temporally. The observed linguistic shifts, including increased use of negation and simpler language at the peak of case surges, are consistent with stress responses. These findings have implications for understanding crisis communication and can inform strategies for more effective public health messaging during future crises. Future research should explore these patterns in other contexts, examine multi-word patterns, and consider the potential reciprocal influence between communication and pandemic dynamics.

While the study included a large corpus of speeches, some speeches were unavailable. The analysis focused on a single level of government in one country, limiting generalizability. The study examined single words, neglecting potential insights from multi-word patterns. The analysis focused on the early months of the pandemic; later responses may differ. Spatial analyses may be subject to confounding factors like population density and political affiliation. Temporal analysis helps mitigate some confounding but not all.