So the gravy train has officially come to an end in the free machine translation space, or at least it has for anybody who has any real volumes! Google announced the deprecation of their API in May 2011 quickly followed by a U-Turn based on the level of “passion and interest” and then earlier this month announced a paid version of their API, which I wrote the article Google Paid Translate API.

What I certainly didn’t see coming, or at least not this quickly, was Microsoft Translator following suit and announcing their own paid version, FAQ here and pricing here.

As I discussed in my last post, the thing I found very strange about Google’s decision was that they have placed a limit on throughput. From what I can see Microsoft have now done exactly the same thing.

It may just be me, but if you are going to launch something commercial surely you want as much volume as possible? Certainly with our own SmartMATE system the principle is the more the merrier!

Since Google Translate and subsequently Microsoft Translator moved to Statistical Based MT, considerably increasing the quality of their output, there has been a surge in use, interest and development around this, with entire business models springing up with both mobile and web based apps. With Google starting to charge I’m sure there was a rush over to Microsoft in the hope that there would be a reasonable stay of execution for all of these businesses – but not so, it would appear!

So what now for people whose business model is built around free translation (including many Language Service Providers who use the service to machine translate and then post edit the output)? Well, if the volume restrictions are deemed to be OK, and businesses have enough margin in their offering, then many will have to bite the bullet and start paying. But for most business models, there is now the need for a massive rethink.

Take Applied Language Solutions as an example. For many years we have offered customized machine translation systems which are very reliant on the customer having enough parallel language data to be able to ‘train’ the system. For many companies that level of data simply isn’t available so we have offered the ability to machine translate through initially Google, and then Microsoft, before post-editing the output. A single project for us could be in excess of the current limits set by Google and Microsoft, roughly seven millon words and 6.4 million words respectively.

Although this is great news for commercial MT providers, like ourselves, it is also bad news in many ways. One of the key differentiators we have over our competitors is that all engines are customized for a specific customer or domain, rather than going for the ‘sledge hammer to crack a nut’ approach. What both Google and Microsoft translators were good at was very broad domain general translation requirements. With both placing limits on the volumes businesses can access they have started the race for MT providers to build massive language models for broad domain applications, with no limits!