I've tracked down the issue: there seems to be some suboptimal logic for the way the code handles wrapping text in this regard. I say "suboptimal" only because if you add a few characters before the offending hyphen it is pushed onto the next line. Alternatively, if you add a remove a few characters the entire phrase is pulled onto the line with the hyphen. This only manifests in the PDF output as far as I can tell.
I'll file a bug on this random, wrapping behavior we're seeing here. Please let us know if it continues to affect your experience.