When a word starts with two vowels with a rough breathing, it's being Latinized with the "h" in the middle of the two vowels. For example, εἷς becomes ehis).
We should write a script that goes through vendor/lsj.json, finds all of the words that are incorrect, and changes them.
Unfortunately, we can't just look for words where the Latin version is vowel-h-vowel because there are some weird words where that's the way they should be, like κρᾱἅρα. I think we need to look to see if the exact match (m in the JSON) starts with a diphthong with a rough breathing and change the Latin based on that.
