Integrity Languages

Blog

Tag Archives: event interpreter

Why Speech Translation Apps Fail

By: Jonathan Downie    Date: October 25, 2018

The Events Management world loves technology. From holograms to RFID badges and from giant robots to automated registration, everyone is on the lookout for some new piece of kit that will drastically improve delegate experience.

In this world, it’s no wonder that machine interpreting/speech translation apps and devices are generating excitement. From the one-way only ili to the shiny in-ear devices from Waverly Labs, millions of dollars have been thrown the way of companies willing to push the boundaries of language technology. Given their slick publicity, even some event managers have got caught up.

Yet there is one small problem with every single one of these devices. They might generate great results in lab tests but, once you take them out into the real-world, the problems quickly emerge. From Tencent’s public fiasco to Wired UK discovering the disappointing truth about the performance of Google’s speech translation pixel buds, it seems that the hype around these devices bears no resemblance to their actual performance.

Why?

Machines Don’t Actually Interpret

What the tech companies don’t tell you is that none of them actually understand what interpreters do. Their devices do three very clever things: they turn speech into text in one language (voice recognition & transcription), then turn that text into text into another language (machine translation), then turn that text into speech again (voice synthesis).

It’s all very smart but saying that interpreters do voice recognition, translation and synthesis is like saying that pianists hit piano keys, press pedals with their feet and follow a rhythm. It’s true but it misses out huge chunks of the process.

Despite all the great shifts in machine learning, the biggest weakness in all speech translation and machine interpreting is that it is not interpreting. It’s not that the technology is bad; it’s that it’s not actually doing the task it is sold to do.

So what do human interpreters do?

Explaining what human interpreters do would take an entire book but there are a few things that buyers need to know about human interpreters. Here are three fundamental ones.

  • Interpreters deal with intonation

Take the English sentence “I only told her I loved her.” That sentence can have seven different meanings according to which word is stressed. In many languages, that would lead to seven different ways of  interpreting that same sentence. Multiply that across an entire speech and you can see that missing the intonation in one sentence can throw off the entire meaning of the talk.

For the moment at least, machine interpreting/speech translation can only deal with the words that are said. Zero intonation and stress data are passed from the voice recognition engine to the machine translation engine. Any meaning that is not found in the words on the page is lost.

  • Interpreters deal with status and number

This might be hard for monolingual English speakers to understand but there is no one way of translating or interpreting “Please sit down” into French or German or any other language that has more than one word for “you”. A human interpreter would know immediately whether you are addressing one or many people and whether there was a difference in social status or need for politeness that would determine which word for “you” they should use. Machines simply don’t know and can’t know.

That might seem like a minor issue but, in the context of say, trying to sell products or negotiating a million-pound deal, the minor annoyance and rudeness of getting that wrong can make a difference. Would you want to take the risk?

  • Interpreters deal with context

This sums up every other point but deserves its own emphasis. Research into real-world interpreting keeps on showing just how many decisions interpreters take because of the context in which they work. They explain culture-specific terms, unknot misunderstandings, shift the language they use according to the needs of the audience, and make smart decisions as to how to deliver the message to the audience. And those are just the differences we know about!

Any machine interpreting system that misses out on intonation, social status and context is doomed to failure because those aspects are just as important as the individual words said by the speaker.

 

These new devices absolutely represent a step forward in technology and might just replace venerable old phrase books in every traveller’s back pocket but for your next event, you should definitely choose humans.

 

If your next event involves more than one language, it makes sense to get help from an expert to get exactly the right interpreting. If you need a consultant to make sure you get great interpreting  when it matters most to you, send me a message so we can chat through the results you want and how best to get them.