[Vulkum]

This might be a stupid idea, but maybe use pairs of translator to get the text translated forwards and then backwards to the original language, and compare how much of the message has been lost? I think this way you can also get a hint of how that native person would interpret the translated message.

[coob]

I really hope it's not a stupid idea, because that's how we translate our product :) - http://utalk.com

We have a web app that two translators use initially independently to translate our entire corpus. The app then asks them to select the best option where there are mismatches (they can also enter a third alternative). There are a couple of rounds of this.

Finally an editor has access to all rounds of the translation in an editable table (http://i.imgur.com/zPJp0sz.png) where they can see who has inputted what, and wether things have been marked as 'acceptable' or 'completely wrong'.

This works well on most of our short/simple strings.

It also helps that we have two voice actors who are also native speakers come in and record the script, where any further errors or better translations can come up.

We do this because we're translating with different goals than the majority of translation purposes. We need to focus on translating for speech at relatively low volume and for those learning a non-native language. Most translation shops focus on written language for native speakers, high volume, speed and cost.

[mrborgen]

Nice, that tool looks like something more startups could use when translating their products. Have you built it yourself?

[coob]

Yep, very much hacked together, but it works well for us.

[mrborgen]

That's not a bad idea actually. This will reveal their quality quite a lot. However, it will be difficult to know which of the translators fault it is if the message gets lost..