Combining linguistic data and behavioral sciences, we use NLP to implement machine learning and inspect social media in a suicide surveillance system. We have conducted prospective studies to understand the linguistic expressions, and discourse features of suicidal subjects. The goal of this research was to build a machine learning processing using the linguistic characteristics. To achieve this, we applied machine learning classifiers on linguistic data captured from heterogeneous sources (blogs, websites, forums, social networks, etc.). The captured data were then used for training machine on information extraction in order to identify linguistic markers of suicide. In this paper, we provide an overview of the automated tracking and monitoring system for suicidal ideation and risk, which draws on predictive linguistics methods and techniques, based on a large sample of suicide messages posted online.