New automated interpreting service ‘not a leap forwards’

The launch of a new telephone interpreting service, which is completely automated and doesn’t use human linguists, is not the step forwards it may seem, a report claims.

The Associated Press reports that Israeli start-up firm Lexifone launched a real-time translation service for phone calls last month and claims it allows any two people to have an “effective conversation” in two languages. Real time translations in English and a choice of seven other languages are available.

However, Max J Rosenthal, writing for the Associated Press, says that the service will not replace human interpreters. He notes that his publication decided to test out the technology and concluded it has a long way to go before it’s a real threat to professional interpreters. “Star Trek’s universal translator will have to wait,” he claims.

Mr Rosenthal notes that the Associated Press tested Lexifone on reporters based in France, China, the US, Mexico and Israel, but that the results were “frustrating”. Not only was the quality of the interpretation – which is spoken in a computer-generated voice – questionable, but it also took much longer to get through the phone call.

The tests found that the technology worked best when used to translate conversations about business, but that when the speech was more informal it struggled. Not only would it take longer to translate the content, but the result would be incorrect. Because of this, the reporters concluded it was best to reserve use of the programme for emergencies, but not to rely on it every day. They also advised avoiding it if the content of the conversation was vital, as “they wouldn’t trust the translations they were hearing”.

Unlike Google Translate and many other free web-based machine translations, Lexifone works by analysing the contents of the conversation to produce the most reliable translation and it is able to understand a variety of topics. Google, meanwhile, uses statistical analysis to work out which translation is most commonly-used for the content it has received. This is because there is so much going on; the conversation is transcribed by the computer, put through the translation tool and then converted back into speech.

Both of these options have pros and cons. Machine translations are fast but they can also be inaccurate, while the Lexifone version is better equipped at understanding context but takes longer to relay results.

Telephone interpreting is a useful tool to have in business, as digital technology makes the world feel an increasingly accessible place. Speaking on the phone can act as a valuable middle ground between impersonal emails and costly flights to meet buyers or investors in person. As many people prefer to do business in their mother tongue, having an interpreter sit in on an international call can actually help form better relations, rather than hinder them.

However, the Associated Press’ tests found that Lexifone struggled with corporate conversations. On one occasion a speaker asked in Mandarin “what’s the issue?” and the English translation came back as “Australian pig”. Mistakes like this will certainly hinder good relations.

Business Day recently reported that author, inventor and futurist Ray Kurzweil had predicted that machine translators would match the skills of humans by 2029. This is still a long way in the future though, which may explain why Mr Rosenthal does not see the livelihoods of real professionals being at risk quite yet.

Indeed, Jon Ritzdorf, a machine translation expert and employee of Moravia Worldwide, tells the Associated Press that none of the combination of different technologies Lexifone is based on is “fully matured”. Currently, human language experts are employed by the company to listen to calls so they can develop improved translations for the system to use in the future.

Even the company’s vice-president of sales Itay Sagie says he would not rely on the machine interpreter as it is now if the conversation was important, advising: “I wouldn’t negotiate a $5 million deal with this.” Meanwhile, Mr Ritzdorf suggests that in spite of improvements, machine translations will never be completely accurate or perfect, adding: “At least not in my lifetime.”