News
DeepMeta now supports 21 languages!
In our continued mission to make Getty/iStock truly accessible to contributors worldwide, we now support 21 languages!
Getty Images has customers worldwide, thanks to its great versatility in subjects, cultures, and creative styles. On the supply side of this great imagery are tens of thousands of contributors, also geographically distributed across the globe.
By providing a UI in many languages, we hope to lower the barrier for creatives around the world to submit great imagery to iStockphoto and Getty Images.
Language Selector
Why these languages?
Easy. We pulled up the list of most common languages in our website statistics and took the top 20 (in addition to American English, the site default). We already knew we have a ton of fans in Thailand (สวัสดี!), but now we've discovered a new fanbase in Indonesia (Halo!). Adding new languages is easy now, so let us know if you feel left out.
For those interested: some history
DeepMeta v3.1 went truly international in August 2017, supporting 8 different languages for its User Interface, later expanded to 10. Making software international is specialized work and requires dedication and patience. We've learned a lot from the localization system used in v3. The main challenge was the bureaucracy involved in obtaining the translations. More specifically, it was the updates that were a challenge.
Whenever we added a new feature to the software, changes were needed across the codebase. We then had to extract the new or changed phrases and send them off for translation to our localization service. This consisted of a network of freelancers, who would then each (one translator per language) be handed a handful of phrases to translate. For the translator, these came out of context. The income they got from this did not justify spending time to get reacquainted with the software and its specific terms, resulting in inconsistent translation of commonly used terms. Next, the translations would come in by email one by one, and we had to refit them in the software. Sometimes the translator did not respect the technical requirements, and changed tags that were not supposed to be changed.
It would be fair to say that this actually posed a brake on the amount of UI rework we wanted to do for a feature. As soon as a feature was ready, we wanted to share it with our users, but the translation roundtrip had to be done first, breaking the momentum.
And this is only the software side. Changes to documentation and website content were even more of a challenge. How do you detect that 2 paragraphs were changed and 1 was added to a webpage? How do you communicate this with the translator, without needing them to re-translate all involved paragraphs again?
Enter DeepMeta 4 and AI
For DeepMeta 4 translations we wanted to learn from our experience with v3, and make internationalization a painless process. I think we succeeded! And how... I know... I know, as creatives, we all have an ambivalent relationship with AI. It's threatening to steal our jobs, taking the most fun part of our business (creativity) out of our hands, and yet... Raise your hands, those of you who don't ask ChatGPT a question now and then, and be honest... more and more of us do...
The translation engine for DeepMeta was built by us from the ground up. Not because we think we're better than all those library authors who offer comprehensive localization solutions. It's because the devil is in the details. The challenge is not turning every phrase in the software and documentation into a "translatable unit", though that is a huge task, only palatable to the most patient and dedicated among us. The real problem to be solved is managing the changes and how to deal with them.
A completely new approach
The solution we came up with is what's called a "functional approach". The framework used for the DeepMeta web app, "React", is functional as well: it starts from "state": your contributor data, and all the rest, including the User Interface, is just a transformation of that state. It makes something intangible, like a Sale or Upload, into a visual panel, button, or text. If the state changes, the thing you see on screen changes as well.
We've hooked the translations into this same philosophy. The moment we change a single word in the User Interface or documentation, it will result in 20 new translated versions, incorporating the new word correctly and in context. We've been testing this novel approach for the last few weeks and are absolutely thrilled with its performance.
Very true, of course: AI translations are not perfect. But may we say that the disconnected translations we got back from our translation freelancers were flawed as well? Not to mention the price. We are now effectively translating tens of thousands of words in DeepMeta 4 into 20 languages. Using our original translation service, this would have cost us tens of thousands of dollars. With AI, the cost is lower than $50!
A word of thanks
As a final word, we'd like to thank the contributors who have done manual translations for DeepMeta 3. Indeed, for a few languages we have been contacted by dedicated contributors who volunteered to do translations for their native language. Needless to say, these were 100% top-notch quality, executed with great consistency. Thank you so much! You know who you are. You have made the uploading experience for contributors sharing your language a real pleasure!
Feedback
As with any feature in DeepMeta, planning and developing functionality is our passion, and we deeply care about how it works for you. Languages are no different. So: does the AI translation work for you, or is it a cringy dystopian mess? Be honest and let us know!
Also, if you see some English text popping up that still appears to be untranslated, please report it as well. Thanks!