Machine Translation is as Ready for You as You are for MT
If you think your organization can’t afford high-quality (statistical) machine translation, think again!
In 2008, I wrote an article about the myths of machine translation, and unfortunately, much of the misinformation about what this technology can and cannot do persists until today. Don’t believe the haters and naysayers who tell you that machine translation doesn’t work and never will or that if MT does work, it is only suitable for the biggest corporations.
It is certainly true that some of the best-known stories of successful machine translation implementations come from large organizations:
- Microsoft has been using machine translation technology to translate more than 100,000 knowledge base articles into nine languages;
- The European Commission, one of the largest employers of human translators, currently uses MT to translate more than 10,000 pages per day in more than 50 language combinations.
And it cannot be denied that many companies harnessing machine translation technology today have made heavy investments in machine translation infrastructure following one of two paths:
- Buying a user-friendly commercial statistical machine translation system and having the manufacturer customize it, which is very expensive, or
- Implementing a free statistical MT system and hiring a staff of computer linguists to customize that system, which is also very expensive.
But while statistical machine translation, or SMT, has gotten all the media attention, there has been a very cost-effective alternative: Combining rule-based machine translation, or RMT, with a controlled-language approach to authoring. The problem with this solution was and is that in order to take full advantage of RMT, potential users typically have to control their content at the authoring stage. And since implementing a controlled authoring initiative is challenging, to say the least, pure RMT systems typically produce less impressive results with uncontrolled input than SMT systems.
With SMT being the technology of choice, what’s a company to do that wants to take advantage of MT but doesn’t have the deep pockets of a corporate giant? The answer is DIY (Do It Yourself) MT!
Recently, a number of startups - like Asia Online, KantanMT, and SmartMate - have emerged that offer low-cost, subscription-based solutions for creating and using customized SMT engines. These self-service solutions allow the user to submit text for translation to an SMT engine that was customized based on translation memories, glossaries, and other materials that the user submitted. Once a custom engine is in place, which can take less than a day, the user then buys machine translation and possibly other services from the provider.
The good news is that these DIY MT solutions can offer the same near-human translation quality that traditional SMT systems produce, but at a fraction of the cost. However, much like traditional SMT, DIY SMT requires relevant, clean translation memories and glossaries to generate machine translations of acceptable quality.