Current approaches in Service-Oriented Architecture (SOA) are challenging for users to get involved in the service composition due to the in-depth knowledge required for SOA standards and techniques. To shield users from the complexity of SOA standards, we automatically generate composed services for end-users using process knowledge available in the Web. Our approach uses natural language processing techniques to extract tasks. Our approach automatically identifies services required to accomplish the tasks. We represent the extracted tasks in a task model to find the services and then generate a user interface (UI) for a user to perform the tasks. Our case study shows that our approach can extract the tasks from how-to instructions Web pages with high precision (i.e., 90%). The generated task model helps to discover services and compose the found services to perform a task. Our case study shows that our approach can reach more than 90% accuracy in service composition by identifying accurate data flow relation between services.