Do interpreting apps really work well?

“I think it’s absolutely terrifying,” said my husband, when he tested out the app, and as a linguist, and as a normal human being, I’m inclined to agree.  As lawyers, there must often be situations where such an app would be useful, or even indispensable. The question everyone is asking is when will such machines replace us as human beings altogether.

I have known of the existence of an interpreting app for some years, but somehow, I never got round to trying it out until, one idle Sunday, while watching the cheesy and highly enjoyable Netflix series Emily in Paris, I saw it in use. The eponymous heroine arrives in Paris speaking not a word of the language, clutches a phone in her hand and presses a button to interpret for her. A mechanised voice accurately repeats her statements in flawless French, but she then proceeds to abandon the device for the rest of the series.

I, however, was determined to try it out for myself. So, one Sunday morning, enlisting the help of my husband, I sat down to try it out. There are a number of similar apps, but I am going to talk about the one called Translate Voice and Text. I would welcome any comments on other similar apps you have used.

Translate Voice and Text

On the opening screen the phrase “Good Day” flashed up in a variety of languages.  I was slightly dubious as to the quality of translating that might appear, given that “good day” sounds rather stilted in English, but I must concede that the quality of interpretation that followed was infinitely superior.

Next I was instructed to “tap the microphone to speak.” Underneath were icons for a keyboard and a microphone, and below that language options to select, eg English with an arrow towards German.

“Hello,” I said, selecting my translation language as French and speaking into the microphone. “My name is Susan, and I am here to test out this language app and see if it works.”

There were a few seconds delay and the text appeared on the screen, and an expressionless female voice in French spoke back almost instantly in perfect French, which appeared on the screen in text in both French and English. I tried a longer sentence. It did the same thing, but this time I had to tap the microphone several times for it to repeat the full two sentences back, but the quality of the translation was excellent.

I tried it in all the languages I teach – Italian, then German, then Spanish, and then Portuguese – and in each case the quality of the translation was excellent, although ads flashed up irritatingly frequently, often blocking satisfactory use.

I explored a bit further what it could do. You could get it to interpret a phrase into French, and then choose a different language, and it would interpret the same phrase into that other language. This could be invaluable if you were in a room with several nationalities, as long as the advertising did not spoil things. It would, however, occasionally stop speaking mid-sentence, but then if you tapped  the microphone icon, it obligingly repeat it, this time with the full sentence.

I explored a bit further and was delighted to find that if you tapped an icon looking a little like a pointed bracket, it could send the text to a variety of destinations of your choice, email, WhatsApp, Facebook etc. There was also the option of scanning it, which I would imagine would be useful if you wanted to use it on a video conferencing call.

What I was particularly struck by was the quality of the voice recognition. I spoke slowly and clearly into the mic and it recorded it completely accurately. Then, trying a different voice, I handed it to my husband, and it had no difficulty in interpreting his voice. We then both tried it, mumbling a little and speaking away from the microphone, and it still worked. Interestingly the interpreting voice in different languages was sometimes male and sometimes female, but there seemed to be no logic to it.

My first attempt voice recognition had been many years previously with Dragon Dictate, voice recognition software that was supposed to improve as it learned how you spoke. I became so frustrated with the inaccuracies in the dictation, and it never seemed to improve, that I abandoned it. My recent experiences have been far more creepily accurate. I discover that I am not intending to dictate something, but words appear on the screen unintentionally, including all the umms and oohs.

How do translation apps work?

Why is it that, while computers have been able to do more complicated calculations than a normal human being for decades, they have only recently been able to compete with the accuracy of a human interpreter?

Google Translate and similar statistical translation programs work partly by detecting patterns in hundreds of millions of documents that have previously been translated by humans and making intelligent guesses based on the findings. Generally, the more human-translated documents available in each language, the more likely it is that the translation will be of good quality. But there are often situations where a machine cannot work out the accurate translation purely based on the given information.

Claude Piron, a long-time translator for the United Nations and the World Health Organization, wrote that machine translation, at its best, automates the easier part of a translator’s job; the harder and more time-consuming part usually involves doing extensive research to resolve ambiguities in the source text. Why does a translator need a whole workday to translate five pages, and not an hour or two?

“About 90 per cent of an average text corresponds to these simple conditions. But unfortunately, there’s the other 10 per cent. It’s that part that requires six [more] hours of work. An example of an epidemic which was declared during World War II in a ‘Japanese prisoners of war camp’. Was he talking about an American camp with Japanese prisoners or a Japanese camp with American prisoners? The English has two senses. It’s necessary therefore to do research, maybe to the extent of a phone call to Australia.”

Subtle mistakes can cause diplomatic incidents, and lawsuits. In 1830, confusion arose over a translation of the French word “demander”, which means to ask – much gentler than the English cognate to demand. A mistranslation inflamed talks between Paris and Washington when a secretary translated a message sent to the White House that began “le gouvernement français demande” as “the French government demands”. The US President took issue with what he perceived as a set of demands.

So, what of the future? Perhaps one day as artificial intelligence becomes ever more accurate, human beings will be written out of the picture altogether.

Susan Isaacs is Director of Languages Beyond 2000, teaching classes in French, Spanish, Italian, German and Portuguese, all easily accessible on Zoom. Her students are from many leading law firms and banks including Linklaters, Norton Rose Fulbright, HSBC and the Bank of England. Email susanelizabethisaacs@icloud.com, website languages2000.co.uk, LinkedIn linkedin.com/in/susanisaacslanguages/.  

Image from Languages Beyond 2000.

Languages Beyond 2000 offers fully interactive language classes on Zoom, with the next classes starting on 27 September 2022. Courses in French, Spanish, Italian, German, and Portuguese cost £195 for a ten week course. Contact susanelizabethisaacs@icloud.com and see languages2000.co.uk. Students include participants from law firms Linklaters, Norton Rose Fulbright, White and Case, Reed Smith and Addleshaw Goddard.