See T13501. This can't move forward right now because "LIKE" in MySQL appears to not be combining-character aware, so even if we normalize the ngram indexes we miss the document on the final "LIKE" query.
Once a fix for that is ready, this might be useful.