What are the major factors contributing to ship accidents, and how do these factors evolve in the long term? This study addresses these two questions by leveraging an unsupervised machine learning method named structural topic modelling to identify the causes of ship accidents. The study analysed 2,341 task errors manually collected from 441 reports issued by four government agencies covering a 45-year time span. The results show that the structure of causes of ship accidents remained essentially the same during this period. This highlights the social-material aspect of navigation technology, indicating that the use of advanced technology may not necessarily lead to safer navigation practices, and the interaction between the technology and human agency must be focused on in the bridge management context. Additionally, the computer-assisted textual data analysis highlights pilot-related factors, which might be rooted in the unsupervised and difficult-to-verify handover procedures between pilots and captains, thereby underlining the importance of appropriate piloting regulations.
