Looking back, how has technology changed the language industry?

Twenty years ago, everyone thought that a computer would soon be able to understand, interact, and translate anything. Today we can turn on a light while talking, yet millions of people are still struggling to learn foreign languages. What happened?

Far be it from me to make an exhaustive analysis of the history of technology and languages, but following various exchanges recently with others in the field, I have made some observations.



Technology entered the world of languages quickly

Great efforts were made early on. I remember one program that was running on a diskette which we could interact with, it probably dated back to the 1980s. Thanks to these precursors, the whole field of languages was imbued with technology fairly quickly.

But we went through a phase of stagnation

Finally, after a few years we realized that between a program that could understand simple instructions and a real language, there was a giant step forward. My impression is that in the early 2000s interest in language technologies has declined. Machine translation remained interesting, but language comprehension lagged behind.

Then came technologies such as Machine Learning, and with them the computing power to work on huge datasets finally making it easy to obtain an acceptable robustness in the understanding of the language, despite its imperfections. Nevertheless, if the progress was impressive, language remained a complex matter and most of the actors in the field understood the limits of the technology.

Today, hybrid is the norm

I know a few actors today who rely on a computer's unlimited understanding of human language. Let's take two examples that I know well. Our friends at Authôt who do audio transcription and us at Pipplet.



Authôt uses artificial intelligence to analyze the audio/video stream and transcribe. The reliability rate of this automatic speech recognition technology is 95%. Depending on the accent or the sound environment some elements can be misunderstood. In this case, Authôt offers a hybrid version, with professional proofreading. We are therefore in a hybrid 95% automatic / 5% human. At Pipplet we would be more like 99% human, 1% automatic. 

We do language level evaluation. We provide companies with a validation of a person's language level. It is possible to do without automated models because your phone doesn't always understand you when you ask "Call Mom", so it won't know to quickly validate that you put your 's' in the third person singular in English. Our evaluation is therefore 100% professional and human. However, technology allows us to help our expert professionals to better identify certain errors.

In conclusion, language learning still has a bright future

So the machine does not dominate the language. And while in some cases of use it is possible to automate, I still don't believe in a world where we would speak directly into headsets for instantaneous translation.

You can find the original article here