TranslateMe, an instant translation service in the Neo ecosystem, published an overview of how it is implementing a voting process to further automate the improvement of its machine translation models. Based out of Mauritius, TranslateMe is building a language translation machine learning algorithm based on a proprietary corpus of ten million sentences.

In the near-term future, TranslateMe intends to provide a commercially-viable API, designed to handle conversation-based translation.

Translation validation through community review

TranslateMe is currently training its algorithm through the use of its Telegram app, which can act as a substitute for the official version. Messages sent within the TranslateMe Telegram App are translated automatically so that each user can choose to speak in and read incoming messages in their own native language.

To improve translation accuracy, users can submit corrections to translations and receive rewards in the form of NEP-5 TMN tokens. Currently, the TranslateMe team manually verifies the suggested corrections before they are fed into the machine learning algorithm.

However, TranslateMe is moving towards validating suggestions via a voting process, allowing community members to receive rewards for determining the most accurate translations.

TranslateMe is modeling its voting process on that of Steem’s, which employs a system that weights active participants votes, and distributes rewards depending on vote weight. It is also researching the possibility of evaluating the reputation of contributors and offering corresponding reward bonuses based on the history of the user’s consistency of accurate suggestions.

Moving forward, TranslateMe is creating groups to test various languages as the team releases new updates that require more specific trials.

The full overview can be found at the link below:
https://medium.com/@translateme/translateme-community-questions-37151b2301c1

Tags: