CoinDesk | 2023

Machine Translation of Stories

We built our translation systems in-house on AWS to increase inbound traffic via search.

While CoinDesk does well in the USA and in countries where English is the primary language, we observed that our competition significantly outperformed us everywhere else in the world. In a bid to increase inbound search traffic, we added machine translation of our stories and a language switcher to the site navigation.

To select the languages into which we would translate, I researched:

The countries with the most cryptocurrency held.
The countries with the most cryptocurrency transactions.
The countries that appeared to make up a significant part of our competitor's traffic according to SimilarWeb.

I then specced out engineering work for two themes. The first was a set of translation systems that would let us plug in different translation services such as Google Cloud Translate or DeepL.

We knew from preliminary tests that Google did well with Latinate languages while DeepL seemed to be better for Eastern European and Asian languages. The objective was to have a system that would pick up our English stories, translate them into the other languages, and publish them back to our site in a subfolder structure, with a subfolder URL for each language.

The second theme was to create templates for the static pages, stories, header, footer, etc. that would also be translated. Translating these was a mix of machine translation and an agency that I hired for key terms like navigation items, headings, and to help build our glossary. This work also included making sure that all SEO metadata was set correctly for the translations so that every instance of a story connected with the translations of that same story.

I was pleasantly surprised by how quickly translations were picked up by Google, and we have observed 70-120% growth week on week for every language added. We still have quite a few more to go, with a total set of around 21 languages to be added to CoinDesk.com.