Posts

US Department of Defense Seeks Universal Translator

The quest for a machine that can reliably translate between one language and another has been going on since 1954, when IBM translated the first pieces of text from English to Russian. Obviously, a “universal translator” would be a tremendous asset to the military, especially as they often have a scarcity of competent interpreters.

Machine translation has progressed by leaps and bounds, but a truly dependable “universal translator” has not yet been developed. However, the US Department of Defense is not giving up. Darpa, the Defense Department’s research arm, has requested $15 million from the US Congress for a “ Boundless Operational Language Translation” system. Read more

Google Translate

Why Machine Translation is Not Good Enough

Machine translation that’s good enough to substitute for human interpreters is like the great white whale, sought by science fiction writers, businesses and militaries alike. However, despite all the hype about the latest iPhone translation app and the ubiquity of Google Translate, nobody has yet managed to produce an algorithm that does the job as well as a bilingual human.

A recent article in Slate on the efforts of the US military to develop a machine translation device to substitute for human interpreters in Afghanistan is a case in point. The article describes the results of a 5 year research effort funded by DARPA. The snippet below shows just how well the device performed in place of a Pashto interpreter:

Rachel asked: “Would you introduce me to him?” Aziz failed to understand the machine’s translation (though he does speak English), so she asked again: “Could you introduce me to the village elder?” This time, there was success, after a fashion. Aziz, via the device, replied: “Yes, I can introduce myself to you.”

Unfortunately, Aziz was not the village elder in question. C3PO, where are you when we need you?  DARPA’s speech-to-speech translation system, called TransTac, achieved an 80% accuracy rate by the end of the research project. Obviously, these are not the droids we were looking for.

The problem, as Slate points out, is that computers are great at storing knowledge and making calculations, but they lack the key ingredient of a successful interpreter: understanding. Attempts to add this essential human ingredient by comparing machine translations to human-created translations and by having real people rate translations for quality also tend to fall short. Plus, it’s slow and expensive. As Slate writer Konstantin Kakaes put it:

The difficulty of knowing if a translation is good is not just a technical one: It’s fundamental. The only durable way to judge the faith of a translation is to decide if meaning was conveyed. If you have an algorithm that can make that judgment, you’ve solved a very hard problem indeed.

We couldn’t agree more, and that’s why it’s so important to use trained human translators and interpreters for important communications.

We recently published a further article regarding the pitfalls of MT, you can view it here: