DeepL and disruptive innovation

This post is a translation of a Japanese article by DeepL.

There was a hot discussion around a tweet about using DeepL to write a paper.

みのん

@min0nmin0n

投稿論文でDeepLを使って書いた英文と自力で全て書いた段落を用意して英文校正に出したら、明らかにDeepLベースで書いた方が修正が少なくて衝撃。技術の発達で英語力そのものの壁はどんどん無くなりつつあるので、そもそもの日本語での読み書き、論理的思考能力の重要性がどんどん増している気がする。

10:45 AM - 20 Aug 2020

1911 7028

translation:
When I prepared the English text of my submitted paper using DeepL and the paragraphs I wrote on my own, I was shocked to find that the one based on DeepL required fewer corrections than the other. The barrier to English proficiency itself is disappearing as technology develops, so the ability to read, write and think in Japanese and to think logically is becoming more and more important to begin with.

I see this situation as a disruptive innovation. That is, while the current machine translation output is too rough for those who currently speak English well to use, the vast majority of the rest of the English-challenged population will find it easier to get fluent English through DeepL than to grumble about it themselves.

I belong to the latter. My English is good enough to finally pass the TOEFL screening at CS universities in the U.S. (I can't pass CMU though), and although I can read and write English papers, conversation is challenging for me. But at this level, DeepL generates more natural-looking English sentences.

Of course, DeepL makes some "terrible" mistakes, such as reversing the meaning of sentences or dropping sentences, but these can be corrected by looking at the English text. Moreover, it continues to retranslate automatically as you make changes, so you can keep selecting English snippets that fit your style and have them translated in no time at all. This is quite an extraordinary experience.

On the other hand, there are some arguments against the use of DeepL

早川　智@コロナウイルス対策には正しい情報を！

@francescodamil6

@min0nmin0n ご自身の英語力を高める上でこれはお勧めしません。慣れてくると人間が書いた文章と翻訳ソフトのそれはすぐにわかります。アカデミアで残って行くには、母国でなくても自由に英語が使いこなせるようにならないといけないと院生たちにも話していますが、DeepLに負けるようではまだまだ修行が必要です。

06:28 AM - 21 Aug 2020

1 7

translation:
I do not recommend this to improve your own English. Once you get used to it, you can easily distinguish between human writing and that of translation software. I tell my graduate students that if they want to remain in academia, they need to be able to use English freely, even if it's not their home country, but if they're not as good as DeepL, they still need to learn to use it.

Ryohei Thomas Nakano (MPIPZ)

@luckystrike1984

作文なんてのはね、前後の文脈での単語の組み合わせとか、響きとか、リズムとか、ちょっとしたニュアンスでの方向付けや暗示とか、そういうのが醍醐味なんであってね、それは個人の好みでありアートであり作品なわけだから、機械にそれができるとは思わないし、機械の作品は僕の作品じゃない。

09:23 AM - 21 Aug 2020

0 5

translation:
Composition is all about the combination of words in the surrounding context, sound, rhythm, and slight nuances of direction and suggestion, and that's what I enjoy most about it, and since it's a personal preference, a work of art, and a piece of work, I don't expect machines to be able to do it, and the machine's work isn't my work.

It is exactly the opinion of existing customers in disruptive innovation. For example, it is difficult to go through DeepL when discussing directly at an international conference, and it interrupts the flow of the conversation. There is a session time limit too. Besides, machine-translated texts are dull and boring. It is more important to improve one's overall English ability. This opinion is correct, and indeed they must have survived that way.

However, the situation is changing. This is due not only to internal factors, such as improvements in machine translation technology, but also to external factors, such as the spread of COVID-19, which led to the cancellation and postponement of several international conferences earlier this year. On the other hand, we started to see some online conferences in the summer¹. In this case, voice recognition technology can be used for Internet-based conversations.

For example, this year's interview for Google's employment examination was an English coding interview using Google Meet, and I remember that the function of transcribing the other person's speech in real time was very useful for me to understand English. The same goes for the YouTube subtitle feature. While it's hard to listen to international conferences and tech conference presentations in real time, it's still easy to read the subtitles as they come out. What would happen if we were able to put a translation in the subtitles? I expect that not a few people will use it.

With the advent of DeepL, the acceptance of machine translation has reached a new level. The chances of seeing opinions like the one in the opening tweet have clearly increased. I'm sure they will continue to use DeepL. I believe that more and more new English learners will rely on DeepL. The flood of machine-generated English is bound to grow. As skilled English speakers increasingly interact with such English, we can expect their own standards to change. Perhaps one day, a sentence that is suitable for machine translation will be considered good. After all, English is a language with more non-native speakers than native speakers. The demand for non-native speakers will stimulate the machine translation market, leading to improved translation performance. It will attract non-native speakers and stimulate further demand. Will we ever see a day of disruptive innovation where machine translation outperforms native speakers?

1. There is a paradoxical concern that the normalization of international conferences online may lead to the anti-globalization of participants due to time zone issues. Attending a 4 a.m. session is an experience I don't want to have again...  ↩