EthereumL10nService

Screenshots

Problem Statement

Slides: https://docs.google.com/presentation/d/1MY-ITYuZAhWVfDiV-v361k3Dp8azV63oPqdUxsl-ByM/edit?usp=sharing Loom video: https://www.loom.com/share/6a705d1c673b4b6c92c7857cd31caa34?sid=1274478b-6c7b-4299-af17-6d261344ff00For Etheruem to reach next billions of users, we will need to break down the language barrier. Only 16% of the population speaks English while 60%+ of the content are in EnglishWe have talked to different communities and identified challengesAI tools are often biased and centralizedCrowdsourcing translations Systems often get abuseNeed ways to Reward Contributors for retention and engagementThus, We're building an incentive network for AI translation agent and human attestorsIn our demo, we translated a video from the Pragama Denver Playlist on Youtube. This makes use of OpenAI Whisper for ASR onto original English transcript, then prompt to OpenAI LLM for the final translatied (Traditional Chinese) file in vtt.Vtt subtitles files are uploaded to IPFS for process and execution of decentralized AIThen with a Chrome extension, we can see the translated subtitles, and end-users can feedback by vote up/down like in StackOverflow. These votes will create crypto-based attestations with user's wallet and send back to our servers.With that we're able to avoid abuse, further calculate incentive allocation base on contributions.We derive contribution base on a few factors such as edit distance between previous translations, time of translation to determine contribution. e.g. Translate early, get more, fix a minor typo get less.Then we issue the final "rollup" contribution attestation using EAS/Witness and commit that on-chain.For incentives, we integrated Grantstack from Gitcoin so project owner can fund raise for their project. Then with allo protocol, payouts can be made accoridngly to contirbution in a trustless manner.We created a custom Allo strategy (ELSStrategy) to check against such attestation before receipent cam ne registered to the protoype.In future we might have deeper integration with on-chain Quatric voting mechanism such that each donor or project owner of project can reinforce who will get the final reward.All in all, this protocol aims to create a DataDAO such that we could accumulate human attestation data as supervised dataset, to further improve our models and expand localization capcity to more projects.

Solution

Architecture OverviewFund DistributionAllo protocolwe created Custom Strategy (ELSStrategy) which check against contribution EAS is attested and not expired before recipient can be registeredContribution/allocationwe calculate score of each contribution via translation / attestation to derive final allocation and register to alloGrant PortalGrant StackChrome ExtensionSolidJSvideojs-vtt to parse and render subtitlesTranslation pipelinesOpenAI WhisperOpenAI LLM prompt for translationsWeblate (Translation Management Systems) to let user to modify subtitlesAttestationsEAS (Ethereum Attestation Service)WitnessAI PipelinesWe run data pipeline to download video from youtube, use ASR OpenAI whisper to output original English transcript. Then prompt to OpenAI LLM for the final translatied (Traditional Chinese) file in vtt.This repository contains packages ofChrome extensionend-user can make use of rendered subtitles at translated videos on youtube, streamETHend-user & attestors can attest translationsSubspackage to parse, translate and deliver subtitlesAttestationsCreate attestatiosn with 2 EAS Schemasattestation on each translationproject attestation on contribution/allocationdistributionspackage to derive allocation base on contributionweblatepackage to upload transcript / translaion onto weblateAPIsimple nestjs api to persist attestationcontracts (Customized Allo strategy)This contract implements Allov2 strategy ELSStrategyGrant PortalWe have https://github.com/0xl10n/grants-stack which is a fork ofgitcoinco/grants-stackCreate & Manage Rounds to fundraise for l10n effortEnd-user Explore available Rounds to donate