Mashing up automatic translation with OCR or voice recognition is like adding another link in the popular childrens game of telephone, which illustrates the errors created when messages are passed down a line.
The cellular world has recently set the internet abuzz with news of some mobile branding translation blunders.
Similarly, Apple's iOS personal assistant application is called “Siri,” a word that means nothing in Japanese, but sounds — when pronounced by a native Japanese speaker — like the Japanese word “shiri,” meaning buttocks. Apple's decision not to release Siri in Japan may possibly be related to this unfortunate homophone.
These brand names, like so many others, obviously do not translate well. However, in general, smartphones and other mobile devices are becoming increasingly useful to conducting business in our multilingual world.
International business travelers and others can make particularly practical use of machine translation when mobile devices integrate this technology with speech recognition or optical character recognition (OCR).
For example, the Google Goggles app can scan a French menu item reading “PatÉ de Canard,” and — if it successfully recognizes the text and internet access is available — will send that text to Google Translate to reveal the English translation of “Duck PatÉ”. Similarly, when Google Translate integrates with voice recognition for “conversation mode,” two people speaking different languages can actually carry on basic conversations, as demonstrated on YouTube by English-speaking girls who used an early version of this function — albeit on a PC — to order Indian food over the phone in Hindi.
The availability of wireless Internet makes these mobile translation applications all the more successful. The power of the translation engine is no longer limited to a mobile device’s hard drive size and processor speed because the translation can be processed in the cloud. Likewise, this connection to the cloud enables users to select the best of multiple translation options or to contribute better translations — a manner of real-time, crowdsourced quality control.
However, like any machine translation technologies, these mobile implementations have not created the miraculous science-fiction cure-all we see in the universal translator on Star trek or the Babelfish of The Hitchhiker’s Guide to the Galaxy. In fact, mashing up automatic translation with OCR or voice recognition is like adding another link in the popular children’s game of telephone, which illustrates the errors created when messages are passed down a line.
If the original language is not recognized correctly with OCR or voice recognition, machine translation will only further garble the meaning — hence the reason why paid voice and OCR applications, like those developed by Nuance, are often a better choice than freeware. We see this truth further illustrated with the amusing TranslationParty.com website, which repeatedly translates text from English to Japanese and back again, each time adding a slight distortion to the result. Each stage added to the process will increase the risk that the final result is inaccurate.
To improve the likelihood of accurate translation, some developers restrict mobile applications to translation of text in a specific domain or limited set of phrases and then train the application to do well with only that controlled vocabulary. For example, the Phraselator uses separate plug-in modules focused on medical or military phrases.
Niche devices and apps apply even greater limits to restrict uses to communicating only a handful of ideas or emotions. Examples include the Bowingual dog translator, the Meowlingual cat translator and the Cry Translator iPhone app for babies. Unlike Google’s 2010 April Fool’s Day prank, these are somewhat serious applications of extremely basic translation technology. Scientists will likely use something similar as they develop a dolphin translator.
Regardless of what apps are installed, all mobile phones place us a mere phone call away from human interpreters. Sometimes the best mobile language solution is to simply call an over-the-phone interpreting service like Language Line or CyraCom.
Other devices attempt to facilitate multilingual communication by replacing many words with images and symbols. The Phrazer handheld medical communication device, for instance, enables patients to touch images on a screen to indicate where they feel pain, and what type of pain they are experiencing, all without verbal communication.
As with any automatic translation, there are appropriate and inappropriate uses for these mobile translation devices. We would be wise to let these technologies help us when the risk is minimal and avoid them in situations where they might cause more damage than they are worth. When machine translation blunders inevitably cause problems due to misuse, the user — not the technology — is responsible.