Many organizations use business policies to govern their business processes. For complex business processes, this results in huge amount of policy documents. Given the large volume of policies, manually analyzing policy documents to discover process information imposes excessive cognitive load. In order to provide a solution to this problem, we have proposed previously a novel approach named Policy-based Process Mining (PBPM) to automatically extracting process models from policy documents using information extraction techniques. In this paper, we report our recent findings in an important PBPM step called task identification. Our investigation indicates that task identification from policy documents is quite challenging because it is not a typical information extraction problem. The novelty of our approach is to formalize task identification as a problem of extracting relations among three process components, i.e., resource, action, and data while using sequence kernel techniques. Our initial experiment produced very promising results.