Anne Merritt at The Telegraph asks the question, How reliable is translation software?
Merritt describes translation software from Google and Microsoft, noting that they both "use a type of computational linguistics called Statistical Machine Translation. This method uses vast databases of human-translated documents...to track patterns in the translations and draw parallels between traits of the languages. Rather than a word-by-word dictionary translation, this technology looks at word combinations and common patterns, so it can estimate the conjugations and structure of a text. In theory the more translation data it obtains, the more refined its pattern-tracking can become, so the translation software is constantly improving.
The author points out some flaws, such as the systems’ inability to catch the many common phrases found in most languages (like “hard to pull off”), and also that they “aren’t sophisticated enough to accommodate the syntax of longer texts.” That’s not to say they aren’t impressive, being particularly good for short texts – directions, tweets, and such. But she points out that they can be less useful and even risky when they’re used in situations where texts are longer and details are more critical, such as at work or in academia.
Imperfect translations can offend or confuse, she notes, with or without the help of technology. As proof, she points to some high-profile and very amusing marketing flubs -- such as when Pepsi's "Come Alive" campaign was translated in Chinese ads as "We bring your ancestors back from the dead."
You can read the entire story here.