It entirely depends on your attack model. A 16-character word is likely to resist brute-force, but might fail immediately against dictionary attacks. How do you define the strength of that password? There's no single answer.
With the scheme you mentioned, a brute-force attack would be largely infeasible assuming your password is long enough. Even then, it's entirely reliant on whether the attacker is targeting your particular Unicode range, or if they're just going for ASCII.
In general, attackers in the "low hanging fruit" model are likely to guess passwords only in the ASCII range, because they're the most common. When 30%+ of passwords are in the top 10k list, why bother doing anything fancy?
However, if you're worried about attackers who know your language and are actively attacking passwords using the same model as you used to create your password, then you might run into trouble. Even so, if your words are sufficiently randomly chosen, and one of them isn't strictly a dictionary word (e.g. a word like "pwned") then you may well be safe. It's quite difficult to judge the strength of passphrases, since their security is largely reliant on unknowns such as the dictionary in use and the potential for mutations.
The truth is that we don't really know if they're secure against in-the-wild attackers in general, because we don't have any solid statistics to back up our intuition. The best advice I can give you regarding non-English passwords is to assume that they were English, and ask yourself whether you'd still trust them.
For further reading, we have a few related questions on the site: