Focusing on the annual report tone and listed companies in heavily polluting industries, a unique text analysis dictionary was firstly constructed based on annual reports and social responsibility reports. Then, using data from listed companies on the Shanghai and Shenzhen stock exchanges, generalized additive models(GAM) and Logit regression were applied to explore potential connections between annual report tone and the severity of violations, as well as different types of violations. The findings reveal a non-linear relationship between annual report tone and the severity of violations for listed companies in heavily polluting industries, which can be specifically divided into three stages. Additionally, there exists a significant linear correlation between annual report tone and specific violation types(e.g., fictitious profits, false disclosure), with certain tone features potentially serving as tools to conceal violations. Therefore, regulatory agencies should strengthen their regulatory measures based on text analysis. For investors, non-financial information deserves increased attention.