I'm a little skeptical as to how accurate this is. It definitely looks like it operates at a better price point than Google's ridiculously expensive Translate API, but does it really offer a huge boost over say, OpenNMT? You can deploy OpenNMT at scale pretty cheaply using Amazon's own EC2 Load Balancer so I'm not sure...

Also do AWS credits apply to this? I and most startup founders I know have upwards of 30k$ in AWS credits usually sitting around so this would definitely go a long way as to integrating translation without running a pre-scale cost.

Ah, these rest API's for translation are amazing. I did a localization from English -> French for a bunch of set top cable boxes by just piping all the UI strings into Googles translate API. No QA, nothing, straight into production lol!

Depending on your needs/budget it's probably not a bad way to go... if you have a large-ish open-source project and want to localize to a few languages, an initial pass like this, while taking pull requests is probably the best way to go.

Yes, the best way to get the correct solution to your problem is to post wrong solution on the Internet and wait for people to explain why you're utterly wrong.

If you didn't do QA, can you really be sure the results were amazing?

if your standard is not "quality", then you don't need to do quality assurance. ;-)

I didn't know Amazon also had a translation service. I wonder how it compares to Bing, Systrans, and Google?

Add Deepl to the list

As an undergrad in French, German and Russian, Deepl has been amazing. Always used linguee to see stuff in context, but deepl is very impressive!

Frupr as well.

Azure too

It would be fun to see a website that lets the user input a string and compare all these APIs.

throw everything on the wall and see what sticks...

how much of these amazon services are even remotely worth using?

AWS is not an indie maker trying to see what sticks. They literally build what their customers ask for and a $30 billion ARR shows that most bets payoff.

If they don't build their own Translate service then they can't compete in that area with other cloud vendors. If you think AWS services are not worth using, then you don't understand the value proposition of managed services built on top of fully integrated elastic/highly-available insfrastructure.

That's not a bad thing. You don't necessarily need to understand that value proposition. But you're definitely not the target audience for this.

More than a few AWS services lean rather heavily on "being AWS" for their value proposition as opposed to being any good at their job.

They do this because they can, and they can because, as you pointed out, once infrastructure and permissions and billing are set up with one cloud vendor, it's usually easier to deal with a mediocre-to-bad product than with integrating an entirely new cloud or tech stack.

AWS is the new IBM, with all the good and bad that entails.

Exactly. From all my years working at "enterprise" companies, I learned that it doesn't matter how useful your product is as long as you get it sold. This is especially true for entrenched products because the friction involved in moving off a given platform is far too high.

I don't know much about IBM.

But AWS all the time gets new services, the services get better and cheaper and they never deprecate stuff.

Yes, and once you build their clunky, buggy minimum viable product into your infrastructure their technical debt becomes your technical debt, even after the second movers figure out how to do it smoothly and do it right.

How do you think IBM wound up peddling outdated crap for exorbitant prices? Exactly the same way. This is how it starts, modern-day IBM is how it ends.

AWS leadership has certain culture which is built for the change and future. But Anything can happen.

They do execute fast.

On the other hand some services from AWS are downright bad. For instance, Elastic Beanstalk is an operational nightmare. The one selling point is that it's easy to get started with, but from there it seems they rely on ops lock-in to pay the bills for this product. It's slow, clunky, and opaque. There are much better alternatives that don't force you to tradeoff usability and lock-in for initial ease of use.

Remember that some services live their life (peak growth and in top 10 list of management), do their job (stable phase) and then run out of steam (tech landscape change, corporate priorities change). It is natural for folks to eventually move off beanstalk into lambda, ecs and forgate. In it peak, beanstalk was the go to service to set up an auto scaling app from code. People have moved on into containers and serverless now. Not so much about beanstalk being bad.

Then again, ELB was amazing a few years back when the tooling for dealing with AWS/GCP/Azure was not as mature. ELB wrapped a huge percentage of the use cases (webapps) in one or two configs.

Sure, now there's a ton of better/more scaleable/cheaper options, but ELB was great before.

it seems that AMZN's spaghetti has pretty high coefficient of stickiness or may be some secret marinara sauce ...

Someone should really make a version of TranslationParty[1] that lets you choose/swap which translation service gets used.

[1] https://www.translationparty.com