AI Translation for a Better World: Jaroslaw Kutylowski / Founder and CEO of DeepL

Direct Talk

Artificial intelligence translation
is transforming communication!

Launched in 2017,
the DeepL machine translation system

currently supports 29 languages

and provides highly accurate translations

for even subtle nuances.

It has been used by
more than one billion people

and has raised the industry standards
for machine translation

even higher than the Tech Giants.

The CEO, Dr. Jaroslaw Kutylowski,

has developed the product
as the technical director

ever since the start-up phase.

Nice to meet you, too.

We are looking towards a world
which really has no language barriers

where it doesn't really make any difference
of which language you are speaking

or which languages
you have been learning in school.

How can AI translation help
to solve the issues of modern society?

Being at the forefront of the industry,

what kind of future does Kutylowski envision?

AI Translation for a Better World

I think there is a lot of
problems in this world

which stem from the fact that
we do not understand each other,

that we do not understand
each other's culture,

that we have just not grown up
in the same places.

While I'm pretty sure that
not all of those problems can be solved

if we're speaking the same language,

I think if we can at least
communicate in a way

that we can exchange ideas
and make us understand each other,

then this goes pretty far
in solving a lot of those problems.

Cologne, Germany.

The DeepL headquarters is here,

a very long way from Silicon Valley.

The company employs about 300 people,

including programmers, marketers,

and experts handling multiple languages.

I'm a native speaker of Spanish.

My first foreign language was English,

then French, German.

And then I started learning
Polish, Portuguese,

a bit of Italian, a bit of Greek.

So I try to, you know,
move around all these languages.

And I'm always very curious to learn
languages from different language families,

different alphabets,

and very exotic ones.

While AI translation can handle
a variety of different fields,

DeepL provides highly accurate translation
systems by specializing in text translation.

How exactly does it work?

Essentially, you can think of any neural
network a little bit as of a human's brain.

And at the beginning,
this brain is pretty much untrained.

It doesn't know what to do.
It's the same as with our small children

who come to this world and can't speak,

can't do basically anything.

And then we train this neural network

and we train it by showing a lot of examples
of how translations could be done

and as the neural network sees more and
more translations, it learns how to do this.

Every time it does its translations
in a good way,

we're giving the neural network,
or the AI, like positive feedback.

And out of that, this huge mathematical
machinery learns how to translate

and become better and better at that and

that is just common and
works for any language.

So I think, actually,
if we had a language of an alien race,

that could be also something that
we could be able to train the AI to do

if we had a proper set of
translations to teach it to the AI.

That's probably not there!

DeepL is chosen by
more than one billion people

because of the high accuracy
of its translations.

What's the secret behind that accuracy?

What is very important with...

that is the design of how this
artificial brain is connected together,

how those connections are made,

and therefore how the text
is processed through the AI.

And this is like what our researchers
are working on daily

and making sure that
the processing is efficient

and that therefore
the AI can understand what the text is.

And this is playing the most important role
in how the quality advances are generated.

So humans' role in advancing
that is a very large one, yes.

The wide range of use cases

include the legal industry, where large
amounts of text need to be translated,

business applications such as cross-border
e-commerce product introductions,

and public agencies
such as the United Nations.

As one example, the German national
railway company, Deutsche Bahn

has created an internal
translation platform using DeepL.

The company's 320,000 multinational
employees around the world

are now able to communicate effectively.

The most emotional comments come actually
from users who use it for their personal life:

for example, communicate with
their family in a foreign country.

Previously there were instances
where we just thought,

"Oh, we won't write that email because it's
just so cumbersome or we just purely can't."

And now you can just do that.

So, I think in general, probably
the amount of communication has increased,

but also the quality,

and therefore the chances that your partner

whom you're talking to
is going to understand you.

That's only a good thing, I'd say.

If AI learns and evolves,
it could perhaps do anything.

What does Kutylowski think?

The quality and accuracy
of translations done by AI

has been increasing
over the last years definitely.

And this is going to continue.

There is different use cases and for example,
I think for conversations

the AI will be becoming
much better and better.

I think when it comes to
poetry and art in general,

a translation is not just the translation,
it's also art by itself.

So I'm not sure whether it really
makes that much sense to apply AI there.

We want to see a human doing that,

So I think there's probably not going to be
even that much focus on that topic.

As we don't think that we'll be able
to help people with that.

Will AI take away translators' jobs?

I think DeepL is usually applied
in different use cases,

and in those use cases where
you wouldn't ask a translator for help -

all of those situations where
you'd like maybe to read a newspaper

or you'd like to quickly write that email,

but asking a translator is
obviously not feasible for that

because it takes just too much time
and it's maybe too expensive.

And on the other hand,
it's transforming a translator's life.

We have many, many customers
who are professional translators,

and their work is just
so much more efficient.

So it's a lot of human
and machine interaction

there to get the best result most quickly.

How does he feel about their small startup
achieving more than the Tech Giants?

I'm proud!

I think this is a very good example of

how actually competition in the markets
brings a specific field really forward

and therefore helps people.

So I'm happy about
the competition that we have,

but I'm also happy about the fact
that we are out there in this world

and can advance this field so much.

Jaroslaw Kutylowski
was born in Poland in 1983.

He grew up
under the influence of his father,

a professor of computer science.

I was first coding at I think I was ten

and therefore pretty quickly
came to the conclusion

that I want to build things
that actually affects our lives somehow

and which make a difference.

Being placed in a multilingual environment
also helped to guide his career.

I was struggling most
when my parents came to Germany

and I didn't speak any German at all,
and I went to school.

So I had to figure that out on my own.

And we humans can do that. And especially
if we are children, we can do that.

This general understanding
of what those problems are

and how you learn languages and
how important it is has also influenced me.

It was an implicit reason why we started
working on machine translation.

After earning his Doctorate
in Computer Science

from Paderborn University, Germany,

he joined DeepL's predecessor,
a translation website operator,

as the technical director in 2012.

The first version of DeepL was developed
in 2016 and released the following year.

Around 2017,

the neural network revolution coming in

and changing how AI can impact this world.

And therefore it was a combination
of the possibility of doing there,

but also seeing the necessity.

As soon as DeepL was launched,

it made headlines worldwide for raising the
industry standards for translation accuracy.

However, its early days
were filled with difficulties.

I think we've just put a lot of mathematical
work into what needed to be done

and then also making sure,

like DeepL is used by
millions of people all around the world,

so we had to build up the data centers.

This is quite a large logistical problem.

We had to employ lots of people

who helped us assessing the quality of
our translations all around the world.

So finding them,

finding ways of working together with them,
was quite a challenging task.

They overcame the difficulties one by one

and increased the number
of languages supported.

Among them, it was Japanese
that broke through a major language barrier.

I've never seen such a
large actual need for translation

until coming to Japan, to be honest.

There's actually countries where

it's very, very hard to communicate
in the set of languages that I know.

DeepL was eager to meet the huge demand
in Japan for a reliable translation tool.

But it had to face a particular problem.

We had to cope with in the Japanese language,

it is that words aren't
separated by whitespace,

which is common for Western languages,

and which a lot of our systems
have based on at the beginning.

And then we had to rework
that and correct that

to accommodate for the Japanese language.

And kind of makes it necessary for us to work
in a slightly different way with the text.

Overcoming such problems, the company
began providing highly accurate translations

and became very popular.

Japan now uses the DeepL service
more than any other country.

When people say that now they're happy
to communicate and able to communicate,

that warm our hearts at the end.

And we are hearing a lot of those stories.

And every time we hear those,
we are very happy about them.

As AI translation continues to evolve,

will it become unnecessary for children
to learn foreign languages as they grow up?

I don't think so.

I actually think that
learning languages gives,

it's a great tool to train our minds.

And it's the same as with learning maths
at school or learning calculation.

Even though there's perfect tools out there
which can offload the task of us.

It's very, very important that we do that.

I have two children,

and I'm very happy that
they're learning foreign languages

and with every advancements they make,

I'm proud of them.

Does he think the overwhelming dominance
of the English language will also change?

I don't actually have a good answer
for this question, to be honest.

I don't know.

It might be that through tools like DeepL,

the importance of understanding
another language will be less so.

Then maybe more communication can go directly
between Japanese and German, for example,

without us speaking English
as the intermediary language.

But then on the other hand,
if everybody is focusing more on English,

just as this kind of language and the tools
are also great for translating into English,

I think it's going to be hard to
actually change that dominant language

because it's been now so established
for so many years.

Through AI translation,

Kutylowski confronts the communication
challenges faced by the human race.

In its own way,

DeepL is also trying to confront a challenge
the whole world faces today:

war.

We have fast tracked the development
of our Ukrainian language

and have released this some weeks ago

only in order to make people
understand each other better here in Europe

now in this situation of crisis

So I hope with that,

at least we are contributing a little bit
to how people deal with the war.

I do not think that nationalism
will get us anywhere as humanity.

And I think I am looking at
this with growing anxiousness,

actually, and hope that
the world and as people will

kind of get over this phase.

What is the motto he has never wavered from?

So this is our motto:
it's "Breaking down language barriers!"

We are focused on making sure
that everybody can communicate

and that's like
breaking down language barriers!

It just says everything!