Your assumption is not wrong at all.
Formal commands have been stored as choices for the In-built speech recognizer to process commands without utilizing much of your system resources. Moreover the commands recognized by the Engine has higher priorities to general conversations/chats therefore if you mix them both only the functional part of the sentence is utilized and the rest gets truncated.
For an year we've been working hard to smoothen the experience. With every iteration of SVA we push the speech recognition part to the next limit. As for now an experimental speech recognition plugin called Online Speech is available in Syn Store