Modeling Preconditions in Text with a Crowd-sourced Dataset