I've generally found it better at doing the opposite: taking readable, concrete terms and obfuscating them into LinkedInish.

In this study it was randomly generated phrases.