Mar 20, 2017 A Word about Automated Content Translation
In early March, the U.S. Department of Defense announced that it awarded a $10 billion language services contract, splitting the budget among nine firms. The contract’s whopping value is a reminder of how important—and often how difficult—accurate translation is. This holds true in the private sector every bit as much as it does for public entities.
For organizations maintaining content-rich and frequently evolving websites, keeping text up-to-date in various supported languages can be a challenge. Three automation approaches often are used to expedite the process.
Simple Machine-Based Translation
With this approach, translations are performed by software. The software analyzes the text, detects the language, and produces the translated version in the requested target language.
“This is the fastest of the options—for example, some solutions respond to requests in a matter of seconds,” Anupam Dutta, WebINTENSIVE project manager, points out. “And it can be the cheapest—some services presently charge a few dollars per million characters.”
The downside: it is by far the least trustworthy option since it primarily just provides a direct word-to-word translation without context or nuance. “And synonyms can present a challenge,” Dutta points out.
Neural Machine-Based Translations
This method also relies on translation by software, but increases translation accuracy compared with simple machine-based translation because the system learns words, phrases, and rules and uses its understanding of these to facilitate translations.
“Neural machine translation can provide better results through evaluating full sentences, and learning from and selecting words based on context,” says Justin Cameron, WebINTENSIVE manager.
For example, the new Google neural machine translation system is more accurate than its old system. It is 58 percent more accurate at translating English into Chinese, and 87 percent more accurate at translating English into Spanish.
“However, accuracy and nuance of neural-enhanced translation still lags behind the capabilities of a professional, native-language human translator,” Cameron notes.
Prices today can range from $40 to $65 per million characters.
Machine Translations Vetted by Professional Translators
“This approach produces the translations with the highest accuracy, as human native speakers review text and correct machine errors, though, of course, it is the slowest of the three processes,” says Dutta. For example, a 500-word webpage may take anywhere from one to three days to translate, depending in part on the complexity of the language pair.
“English to Spanish, which are relatively similar tongues, takes less time to translate than Korean to Urdu,” Cameron explains.
This also is the highest-budget option, averaging today between 10 cents to 20 cents for each word of a language pair. At times, discounts can be negotiated.
To learn which of these approaches may be best for your website, just call: 212-447-1100.