Hey Daniel,

For that reason, a wallet implementing this convention should not just see “French words” and guess. It should know, or ask, which wordlist mode is being used:

This is the exact problem I was referring to. Imagine your french uncle dies and leaves you a 12 word french seed phrase. You install a BIP39 wallet and import the seed phrase. The wallet asks you: Is this a BIP39 seed phrase or a TZUR-translated seed phrase? You don't know what that means - all you have is a list of 12 french words. The only thing you can do as a user is try both. And if the wallet doesn't ask you at all (because maybe it only supports one format or the other), then the wallet happily imports the seed without asking and you have no idea the distinction even exists.

A seed phrase is supposed to contain all the necessary information to recover a wallet, either explicitly in its encoded payload, or implicitly in its format. That way a user doesn't have to record or remember any other extraneous meta-information - Just the seed phrase.

To make your translated wordlists work like this and fix the ambiguity, I would recommend you consider changing the encoding or the format to ensure a TZUR-translated display phrase cannot be interpreted as a BIP39 non-english seed phrase. So for instance, using a different number of words. Or use wordlists disjoint from BIP39, so that the probability of a random TZUR phrase containing at least one non-BIP39 word is overwhelming. 

One clean way to do both would be to increase the size of the TZUR wordlists. For instance, instead of encoding 12 words with 11 bits per word, you might encode 11 words with 12 bits per word - the exact same number of bits are encoded in both cases. To map the phrase back to BIP39 is as simple as changing how those bits are interpreted. Since a 4096-length word list would be double the size of a BIP39 wordlist, it must include at least 2048 words not present in the BIP39 wordlists, and so a randomly-sampled word has at most a 1 in 2 chance of being in the BIP39 wordlist. Thus the chance of all 11 random words just happening to be BIP39 words would be at least 1 in 2048. You can (and probably should) reduce that probability by reducing the size of the intersection between the TZUR and BIP39 wordlists.

You don't have to use those exact numbers either - as long as your encoding is able to package all the entropy bits from the equivalent a BIP39 seed phrase, it should work. (e.g. 13 words with 10 bits per word, encoding 130 bits). 

You could use wordlists which have size that isn't a power of two, and use some base-x math to reconstruct the entropy as a big unsigned integer. 

You could make your wordlists shorter instead of longer, though you'd need to be careful to minimize the overlap with BIP39 wordlists.

You could use a different word list for specific word indexes, e.g. the first word must be a non-bip39 word, and so it encodes less information but guarantees the phrase can't be misinterpreted as BIP39.

Anyways just some ideas. Hope it comes in handy.

regards,
conduition
On Saturday, June 13th, 2026 at 9:46 AM, Daniel Osemberg <daniosemberg@gmail.com> wrote:

Hi conduition,

Thank you for the thoughtful feedback. I agree that ambiguity is the main issue any proposal like this has to handle carefully.

To clarify the design: TZUR display wordlists are not meant to replace or reinterpret existing BIP39 wordlists. They are separate, index-parallel display wordlists whose purpose is to render and accept a user-facing mnemonic in another language while keeping the canonical English BIP39 mnemonic as the seed of record.

So the derivation path for a TZUR display mnemonic is always:

localized TZUR display words → word indices → canonical English BIP39 words → standard BIP39 checksum validation → standard PBKDF2 → standard BIP32/BIP84 derivation

The localized words themselves are never passed directly into PBKDF2. Only the canonical English mnemonic is.

Your French example is exactly the edge case that needs to be explicit. A legacy French BIP39 mnemonic and a TZUR French display mnemonic are not the same thing. They are two different encodings that may use the same human language, and they must not be silently treated as interchangeable.

For that reason, a wallet implementing this convention should not just see “French words” and guess. It should know, or ask, which wordlist mode is being used:

  1. Legacy BIP39 French wordlist

  2. TZUR French display wordlist

  3. Canonical English BIP39

The reference design also includes stable wordlist identifiers: language code, version, and SHA256 of the exact wordlist file. A wallet can persist that metadata alongside the wallet, and use it during restore to avoid ambiguity. But for maximum portability, the wallet should always show the canonical English mnemonic as the universal recovery form.

So I think your concern is valid. The safe rule is: never auto-detect between legacy non-English BIP39 and TZUR display lists when both exist. Make the mode explicit, keep the English mnemonic available, and treat the display mnemonic as a UX layer rather than a new seed derivation scheme.

Regards,
Daniel


On Saturday, June 13, 2026 at 7:36:05 PM UTC+3 conduition wrote:
Hey Daniel,

So basically this would allow a user to translate an english language BIP39 seed phrase to/from other languages, insured by the knowledge that it be converted back if needed for compatibility with english-only BIP39 wallets.

Neat idea. This is how BIP39 should've been done originally. Today, a BIP39 seed derives a different master key depending on what language is used to encode the source entropy, which was arguably a mistake because it breaks compatibility between implementations written for different locales, and incentivizes everyone to use seeds of the same language so that they're maximally compatible (which is exactly what happened).

I worry your proposal here might cause confusion if introduced today though. Say you are given a french language 12-word seed phrase. Do you map the word indices to english and then run PBKDF2 with your algorithm? Or do you run PBKDF2 on the french version as specified in the original BIP39?

There would need to be a clear way for humans and software to distinguish between a "locale-mapped seed phrase" using your spec, and a legacy french BIP39 seed phrase, so they know how to derive the correct master key.

I guess one could argue that non-english BIP39 seeds are so uncommon that you could safely assume the former, but still it leaves open an unfortunate ambiguity which could lead to lost funds in some cases.

What do you think of this?

regards,
conduition
On Saturday, June 13th, 2026 at 12:04 PM, Daniel Osemberg <danios...@gmail.com> wrote:

Hi list,

I would like to ask for early feedback on an idea before attempting any formal BIP submission.

The proposal is a display/input layer for BIP39 recovery phrases in additional native languages.

The important constraint is that this does not change the BIP39 cryptographic flow. The canonical mnemonic remains the existing BIP39 English wordlist, and PBKDF2 is still performed on the canonical English form.

The native-language lists are index-paired to the English BIP39 wordlist. In other words, each native word maps to the same 0-2047 index as the corresponding English BIP39 word. Wallets could display or accept the native-language form for UX purposes, but internally normalize back to the canonical English mnemonic before seed generation.

Motivation:

Many users around the world are asked to back up and restore Bitcoin wallets using English recovery words, even when English is not their native language. This creates UX risk, spelling mistakes, misunderstanding, and lower confidence during backup and recovery.

This proposal tries to improve multilingual recovery UX while keeping compatibility with existing BIP39 behavior.

This is not:

A new seed scheme
A replacement for BIP39
A new cryptographic standard
A change to PBKDF2 input
A wallet-specific format

It is intended as a display/input convention for wallets that want to support native-language recovery UX while preserving canonical BIP39 compatibility.

A draft implementation and wordlists are here:

https://github.com/osem23/bip39-wordlists-tzur

I would appreciate feedback on:

Whether this idea is appropriate for the BIP process at all
Whether it should be considered informational rather than standards-track
Whether the index-paired approach creates hidden risks
Whether wallet developers see practical value in this
How native-speaker review and normalization rules should be handled
Whether there is prior work I should study before continuing

Thank you,
Daniel

--
You received this message because you are subscribed to the Google Groups "Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bitcoindev+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/bitcoindev/e4ee1a70-aa70-4dbe-9f6c-27c26c5d17e0n%40googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bitcoindev+unsubscribe@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/bitcoindev/7fe5fea8-ae9d-4081-a94f-fa9be0677012n%40googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bitcoindev+unsubscribe@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/bitcoindev/kxuhfn7jqWsmcvm6ANEIYxdL8FiGzgHt5_tx9weHR9WMfIv_wAopjsBK9zbIb_3UdL5sLzzOTkCCdXe0YmfcbKX4_n3q5Fp8HUd4LlfPojo%3D%40proton.me.