basically the predictive text input API is non-public, so you can't access it directly.
Anyway see editor example of how to use editors, with them you can set flags to allow/disable the text input modes and let the frame work handle the actual use of the framework.