RFC: A Web3 Translation Tool

[myz1237.eth]

RFC written by Alex1237 in reference to the previous RFC.

What is it?

A suite of translation tools that facilitate and incentivize the community to create high-quality, localized text resources in Web3.0.

Why do I come up with the idea?

The DAO has a large, rapidly increasing number of wiki pages, forum posts, documents, etc which are written in English. To be truly inclusive, we should translate all these resources into multiple languages accurately and on time.
The current translation process is largely manual and not scalable:
So far, we have translations for newsletters and wiki pages. The current way relies on volunteer work which does not provide enough incentive for the general community to participate in and produce high-quality translations, although we have 7 members in the translation team.
This could be a tool provided by D_D for other DAOs as well.
If we could get the economics right, the tool could even generate income for the DAO and everyone who works on the translations.
If we build it smartly, it may become a public good, widely used by other DAOs or individuals.

How does it work?

Basic Data Flow

A brief introduction to the platform：
Our platform needs 3 roles: translators, reviews, and language leaders. Their duties will be covered in the next section. First, select a source as input, such as webpage, word, notion, or substack. The platform could cut contents into pieces and store those pieces into IPFS or Arweave after being translated through DeepL API. Second, those translated pieces are corrected and reviewed by translators and reviewers, respectively. Next, language leaders handle special cases (illustrated in subsequent sections), and rewards and penalties are executed automatically. Finally, the translated articles will be displayed by a browser extension.

Role Duties

We have 3 roles in our system: language leaders, translators, and reviewers.

1. Language leaders are assigned or elected by the DAO. They whitelist members from DAO communities and add their addresses to the system. Bad actors are removed by leaders if needed. Another assignment for leaders is to set up content to be translated (Leaders could introduce sources through the platform or a browser extension). In the future, leaders could list translation rules for the system (eg, agreed translation of phrases and expressions).
2. Whitelisted DAO members deposit DAO tokens into a deposit pool with a cooldown period, to register as translators. Then, translators choose which article they’d like to work on. The system distributes several translated pieces without correction. After correcting article pieces, translators need to sign with their wallets. Note that translators are not allowed to change pieces or switch articles, once they are proofreading. An X-day deadline is also set for each proofreading progress. Translators who give up pieces on hand or do not finish corrections before the deadline will be punished by deposit deduction.
3. Proofread pieces are collected by the system and removed to a piece pool under review.
4. Like translators, a deposit is a must and reviewers could pick up pieces from the pool without limit to a DAO. In other words, reviewers may work on multiple articles in various DAOs simultaneously and should complete review tasks before the deadline. Any violation may cause a penalty on their deposits.
5. To reduce overheads to leaders, each piece is sent to X reviewers and the system does a similarity test. Only controversial results will be transferred to leaders for final decisions. Otherwise, the system automatically accepts the piece with the highest similarity as translation results. Other similar pieces are tagged as accepted and reviewers get rewards. Eg: 4/5 pieces are similar and tagged as accepted, and the reviewer of 1/5 piece will be punished. If pieces are divided into 2/5 and 3/5, leaders should determine which group is accepted.
6. Leaders receive pieces with problems and make decisions.

Structures

High-level Structure

Our platform could hold multiple DAOs with many article projects. Each leader represents a language, responsible for article projects translation.

DAO Member Structure

Each DAO should assign a leader to a language and the leader manages multiple translation projects.

Treasure Structure

We have two treasure pools: deposit pool and bonus pool. As its names imply, translators and reviews could deposit and withdraw their funds. Any violation will cause penalties. Those penalties from translators and reviewers will be added to a bonus pool DAOs set up with their governance tokens.

Essential Dashboard of each role

Whole progress flow

How to display translations?

Currently, my idea is to show translations using a browser extension, which keeps the contents’ formats.

Possible issues

1. Similarity tests among review results may not work if most reviewers are bad actors. A good algorithm is needed for similarity tests. otherwise, language leaders act as reviewers.
2. Translators deposit DAO tokens into the pool, but which tokens should reviewers use for a deposit, a platform token, or limiting reviewers to a DAO?
3. Reviewers’ abilities. Professional and trusted reviewers could ensure that translations meet s specific level of quality and consistency. Do we need verification progress for reviewers?

Dev Roadmap

Currently, we are in concept definition and discussion step. All suggestions are welcome. I’d like to figure out two questions:

1. Is it worth designing and implementing this complex system? Because i18n is just an optional step of success, it’s possible for our dao to throw translation tasks to commercial teams, though which may be against the DAO’s intents. If not, seeing this project as a long-term aim is an appropriate choice.
2. Does this project make translation complex, any redundant role or step? Now, we are relying on voluntary efforts, not a substainable method for i18n. If so, what does your translation progress look like in your mind?

Welcome to any advice and glad to talk in a meeting if available.

[Joris]

@myz1237.eth

A suite of translation tools

Hey myz1237.eth, thanks for sharing your ideas and concepts of building such a complex unit.

I’ve worked as a BizDev almost 1,5 years for the leading translation company in Germany with over 7000 translators worldwide and internally developed IT solutions for managing the overall translation process (ERP, TMS, Web-Plugins, …). Thus, I’ll try to give you my best 2 cents on your post.

Is it worth designing and implementing this complex system?

It depends. For me personally, I’m seeing i18n on the optional side of things, because developers of this DAO probably are used to work in an English-speaking environment. Also, the crypto branch we’re in, is heavily dominated by English content. That being said, this condition could also lead to an unsatisfied demand for native content which truly support inclusion.

In terms of complexity, I would shrink the scope to make it manageable. The idea of “a tool provided by D_D for other DAO” feels to me like one step too far.

@myz1237.eth

what does your translation progress look like in your mind?

Your outline is more centered around the conceptual interaction of people, as I understand it. I would focus more on the nitty-gritty part of translation and the support technical solutions. You mentioned for example the problem that a translation could not be aligned with the interest of the client of this translation. For this (and other) purposes there are established solution like Translation Memory Systems (TMS) which is a database that stores sentences, paragraphs or segments of text that have been translated before. This is crucial in order to deliver reliable high-quality content with various people involved, because language is in the end highly subjective and context driven.

Hope my impressions are useful.

Best,
Joris

[HenryL]

I think this is a good idea, though not easy to implement.
I’m not sure about other language communities but in the Chinese web3 community (including but not limited to many Chinese DAOs) I can see the no.1 activity in terms of content creation is translation. Even though there are quite strict regulations but web3 is still hot. If you check Mirror you’ll see tons of translated content in Chinese. I personally joined several translation group too. That’s why I think this idea is valuable.
Good to see a professional’s opinion from @Joris . I’ve worked for a localization company (a China local competitor to Lionbridge) too, like 15 years ago, though my role was not in the translation part. I’ve seen colleagues used Trados you mentioned and found it very powerful. But when speak of web3 content translation, one interesting thing is, this kind of translation is pretty “wild”. I mean, it’s totally done by individuals. The typical translation process is just AI translation + human modification. Some big guilds have reviewers too. Even though some of these individuals are professional translators, the web3 translation activities are all very loosely constructed. It never aim at the traditional publishing standard. A “OK” quality always works. Sometimes they even translate in a very oral way. This actually works better because readers find it easier to understand.
My expectation for such a tool would be:

1. Really simple in process.
2. I don’t see much need for the tool itself to be web3, like store as NFT. Who pays the gas?
3. Direct publish on Mirror.

That said, these needs are not strong enough for a web3 tool yet. Web2 collaboration tools seem to work well so far.
I’ve also discussed this a little bit in one of the biggest translation guild in Chinese web3 community, SeeDAO. They have similar thoughts.

[myz1237.eth]

Thanks for your reply, henry. Agree with your comments on translation quality. Currently web3 translation does not focus on precision but readability.

We need strong needs

[carllippert]

Hi!

Cool idea.

I recently built a chrome extension to translate substack articles on the fly leveraging webapis, google translate, and the oxford dictionary api for definitions.

Tool was built to help me immerse in languages reading the content I already consume.

Landing page → https://worldli.me

Dom manipulation to use chrome extention for this requires parsing of each platform individually, similar to scraping each site would have some custom configuration needed ( although there are probably more sophisticated ways to do that )

On the work flow for translators for human translation.

I personally think that work orchestration should be a general system vs a system created just for say translation. I have a personal project related to that, that I probably should create into an RFC since I think its a general piece of internet infra that should exist. Maybe this post is the spur to get me to do that!

Cheers

[myz1237.eth]

An update here:
For payment, I would like to integrate with Dework. Income/rewards member get in this platform will be shown in the Dework dashboard and members get payment through Dework Bunch Payment functions. So our translation tool does not need to risk at assets protection and just focuses on translation and review, aha!

[myz1237.eth]

It does not matter, I think. Our DAO is to build the public goods not a profitable products. Certainly, revenue is better, but not a must