MT (Machine Translation) post-editing projects have been the starting point of numerous debates online recently. If you take a quick look at the Proz forums, you will find many of these passionate talks - the variety of metaphors used there to denigrate MT is a tribute to our community's creativity.

Our profession is constantly evolving, and I don't see the emergence of such projects as a threat to my job or to the translation industry in general. If anything, they represent a growth opportunity, as they could allow more end clients to afford linguistic services.

That being said, most MT post-editing projects I've come across so far didn't seem to make any sense for the parties involved, except maybe the agencies handling them.

Most of these projects fall into the following two categories:

1. Content collection and analysis projects

Many such projects come from clients who just want to have a general understanding of the texts they submit.
Two scenarios here:
- The MT output is more or less "fixable": in this case, is post-editing really necessary in the first place? Readers can probably put back the pieces together and easily spot the odd mistakes putting things back in context.
- The MT output is poor and needs a complete rewriting: Why wasting time with MT if the text will need to be retranslated anyway?

2. Projects aimed for publication

For these projects, the goal is to obtain a translation good enough for publication. In this case, terminology and style are critical.
Bad news: general MT engines have no idea of the terminology and style you want to use. Moving MT bits around, rephrasing sentences and checking terms will end up taking at least as much time as retranslating from scratch.

Specialized MT engines: a promising approach that still needs refining

The only case that seems the promising for the future is that of customized MT engines. If you train a MT system to use your terminology and sentence patterns, you can save the post-editors a significant amount of time. I have seen a few of these engines in action for repetitive and "mechanical" texts (user manuals for example) and I have to admit that in some cases they can be of help, although the time gain generally won't be enough to make up for the rate-per-word cut translators are asked to accept for such projects.

That being said, I can see a real potential here. With further improvements to general MT systems and more tweaks to adapt them to certain specialization fields, they may become extremely handy tools.

Conclusion

As things stand, post-editing general MT system translations doesn't make much sense. They are extremely helpful for people trying to gather information in various languages, and this is really what Google Translate, Bing Translator and their competitors should be used for.

Specialized MT engines, while still a bit too raw to be really helpful, have an interesting potential that deserves to be explored.

Like it or not, MT post-editing is here to stay. It is hard to predict how much time it will take specialized MT engines to come to maturity, but they ultimately will. The best we can do is to prepare for these changes, just like every time our industry faced major changes - otherwise we would still be working only with a notepad and pens. Let's be part of the change and prepare to guide our prospects when they are considering the use of MT.