Recently, simulation platform and libraries are provided from several research groups. However, development of VR-based surgical simulator takes much effort not only for implementing simulation modules but also for setting surgical environment and choosing simulation modules. Surgical manual describes knowledge of manipulations in surgical procedure. In this study, language processing is used to extract anatomical objects and surgical manipulations in a scene from surgical manual. In addition, benchmark and LOD control of simulation modules optimize the simulation. We propose a framework of semi-automatic development of optimized simulator with surgical manuals. In the framework, SVM based machine learning is adapted in extracting surgical information and XML file was made. Simulation programs were crated from XML file using a simulation library in different system configurations.