Since the vast majority of BIP39 wallets supports only the English wordlist, it is strongly discouraged to use non-English wordlists for generating the mnemonic sentences.
BIP-39 since recently also states under shortcomings:
Generated seed depends on the wordlist that was used for the mnemonic. Because the "mnemonic to seed" process uses the mnemonic sentence directly rather than the original entropy, translating the mnemonic to a different wordlist necessarily creates a completely different seed. This is not an issue if you only support the English wordlist, as recommended above.
So yes, you can use anything custom, but you cannot ever lose it. It's not a recommended use because text has encoding of its own, especially when using non-ascii characters. If you make something custom, drop the hash(words) and instead do.
Generation:
byte entropy -encode-> words
|
\-hash-> BIP-32 seed
Honestly? This is a square wheel. So by all means go and reinvent it, but it'll be extremely painful to convince anyone to implement your new standard, which is basically why BIP-39 is used by everyone except electrum.
Not really. It's recommended against in the BIP text
BIP-39 since recently also states under shortcomings:
So yes, you can use anything custom, but you cannot ever lose it. It's not a recommended use because text has encoding of its own, especially when using non-ascii characters. If you make something custom, drop the
hash(words)and instead do.Generation:
byte entropy -encode-> words | \-hash-> BIP-32 seedAnd restore:
words -decode-> bytes -hash-> seed