AI Localisation: Help Me Help You
Today we look at AI localisation. Essentially, how to use AI to reach a broader market at scale.
Its a really simple question.
Does AI Localisation work at scale?
As always, the answer is
yes and no, but mostly it depends on what you are trying to do.
The context is this: localisation project managers often burn themselves out by performing tons of routine operations every single day.
So the question becomes, how to use AI localisation at scale, and then, how to best use AI in localisation efforts at the process level?
Does AI Localisation Work At Scale?
By at scale we mean if
you need to translate content regularly and use translators. And
we narrow our question to media translation specifically.
Given the sheer number of AI’s which can be used for translation, and the number of available languages, the answer is
yes - AI does work at scale and its pretty good.
What Challenges Does AI Localisation Face?
Ok - not really.
Assuming a standard Speech-To-Text AI (for the transcription), or a standard Text-To-Text AI (for the translation), this is how you want to think about it.
(a) if the content is gossip, AI will probably be pretty good, (b) it'll be less good for say medical or legal content, (c) when it comes to quantum physics best of luck!
The higher the information complexity, the worse an AI will perform.
The reason is fairly simple. Complex information is inherently more valuable, and so less common on the Internet. Hence it is very hard to train an AI to work with the complex information.
How To Use AI On Your Projects With Confidence?
Use AI to assist your people, not replace your people.
The workflow you should be thinking about is:
- Upload your content and transcribe: Use AI to do the first pass transcription and a human subject matter expert to post-edit.
- Upload your content and translate: Use AI to do the first pass translation and a human subject matter expert to post-edit.
How Does VideoTranslator Simplify The Post-Editing Process?
So this is what you want to do:
- Pop out the video to PiP so you can easily scroll as you make edits.
Try to make nice sentences - no one want to read a single word fragment.The way to think about Open Captions is like so - you want nice 4/5 word sentences which are easy to read, not single word like “on” as shown in the image.
Time fragments should be a minimum of 1 second long, ideally 1.5 seconds or so,to make it easy to read quickly.
The number of edits you make is called
the edit distance, or edit delta, and it refers to the number of changes post AI before a finished product is ready.
We use the VideoTranslator to add transcripts to our video content.
The key points are:
Use the AI as a first pass before your subject matter expert to increase productivity.
AI's are best used to simplify and speed up workflow. They should not primarily be used to replace staff.
AI will fail in localisation without the use of subject matter experts, especially as your content moves up information complexity.