For Minnan: The first link provided by
@audreyt under "collection" doesn't seem to work (it transports you to a page containing a poem). There is another one under the parsing area which does though (
https://github.com/g0v/moedict-data-twblg). The dict-twbl.json file in that directory looks usable, although the documention seems to suggest that there are more up to date versions which I can't find.
For Hakka the
http://www.audreyt.org/newdict/hakka.tar.gz ultimately contains a folder with individual html documents for each entry; don't know how easy that is to work with.
For Minnan at least, one would also have to decide how to deal with non-standard-Unicode characters such as 亻因. Many of these are supported in the newer extensions (C and later), but a few are not. For Hakka I can't say how big this problem is because I can't speak Hakka.
Btw
mikelove, is it possible to define two fonts for characters, one as main and the other as fallback in case main doesn't support a character? The reason I'm asking is because I have so far been unable to find a single font which includes Unicode including the newer CJK extensions. Usually there only is an extending file which contains
only the extensions. So in order to display a text which contains both "normal" and extension characters, you have to switch font for the new ones.
EDIT: Just discovered that the list in the json file for Minnan seems to use combining diacritics instead of hardcoded marked vowels (e.g. ā (a + combining diacritic ¯) instead of ā (hardcoded ā) for the letter a with a macron). At least if I copy and past the code into a text editor, I can delete the diacritic and the letter stays. I guess that would greatly simplify the program needed to convert the romanization with diacritics into romanization with numbers.
