Computers Are Bad At Voice Translation

Bears are so smart they probably know that relative to humans, computers are bad at voice translation. Some bears might even translate languages better than most computers. Though they’re pretty shy in the wild, so it’s unlikely you’ll encounter one that translates the specific languages you have in mind for your telecom application, IVR system, or even for an auto-attendant. For on-hold messages, they’d likely do a fairly decent job since meaning and intent are more important than exact ideas. But for a menu-driven application requiring voice prompts your best bet is to have a qualified human translator do the job. How would I know that computers are bad at translation?

computer translation bad
Does this look like a pound sign to you? Here’s a pound (hash) sign for reference, #.

Some years ago I was working with a German voice talent and in the middle of voicing some fairly standard IVR prompts, towards the end of a prompt, she burst out laughing. I asked her what was so funny and she said her script read,

“Push the symbol of the bear.”

Because our talent didn’t have a copy of the English script, she wasn’t sure at first what the original, intended verbiage was supposed to be. We stopped the session and found the English copy for comparison. What the client wanted was one of the far more generic professional voice prompts, “Press the pound key.”

This is one of our more shining illustrations of why using a computer for language translation is problematic.

See? Voice Translation By Computer Can Be Bad

Using a computer to translate one language into another is a common practice because it’s “easy”. Often, native speakers of the output language are not always readily available, or are perceived to be expensive – especially when there are free computer translators on the internet!  And yet, the above example demonstrates that fast and cheap doesn’t always deliver the desired result, let alone a coherent one.

We see time and time again how the nuances of language – grammar, context, current vernacular – are not something about which a free computer translation service can make a good contextual judgment. In case more convincing is needed, let’s push some verbiage through a popular, ‘good’, free translation service (in other words, a computer) and see what comes out of a back-and-forth, English to Spanish, Spanish to English set of translations.

Here’s the Spanish:

“No utilice el ordenador para hacer las traducciones. Contratar a un humano en su lugar. Ahora, sólo para una prueba, tanto a través de una popular, ‘buenas’, servicio de traducción libre (en otras palabras, un ordenador) y vamos a ver lo que sale de una y otra vez, inglés a español, Español a Inglés conjunto de traducciones.”

And the same, back to English:

“Do not use the computer to do the translations. Hire a human in its place. Now, only for a test, both through a popular, ‘good’, free translation service (in other words, a computer) and we’re going to see what comes out of time and time again, English to Spanish, Spanish to English set of translations.”

It’s not all wrong but it’s also not all correct. When read aloud, it just doesn’t sound right.

Now imagine this kind of mangled content representing your company in emerging markets. In fact, try it out for yourself. Take some of your marketing literature and translate it via machine a few times back and forth and see how it can evolve into sometimes subtle, sometimes complex, sometimes bizarre, and down-right unintelligible gibberish.

Real people are good at voice translation for IVR and Call Center applications, whereas computers are really bad at voice translation

We recommend engaging a human for translation. Unlike computers with rigid rules and a limited database set of interpretations, good human interpreters know that translating the word-for-equivalent word of a script may not convey the true meaning of the original speech. Idioms, colloquialisms, and even single words have shades of meanings that need to be considered in the context of the whole, at which native speakers still excel over automation.

As local resources often prove time-consuming to discover, vet, and coordinate, consider going with a company that has access to dozens of native speakers who do voice translation. Talk to us if you have questions about your next translation and recording project.

And share with us some of your odd computer translations of your own literature in the comments!


–        Written by Dan Nelson

Leave a Reply

Your email address will not be published. Required fields are marked *