So you don't care about validation, as long as it does not affect display (something that is often unpredictable). Well, it's your book and your choice, I won't go further into this.
But if the reason is being able to identify Greek text, I would suggest using a class or, better yet, an xml:lang attribute (although it's not properly handled by ADE). This would allow to identify the Greek fragments, and also to style them in a different way (like using an embedded or different font). When xml:lang is recognized, it also allows the reading app to use different hyphenation or text-to-speech patterns.
What I mean is something like this:
Code:
unmitigated villainess. “<span xml:lang="grc">δεινὸν τὸ τἰκτειν ἐστίν</span>,” she says