I have received a number of comments on an article published on opentheword.org two days ago in a strange language.
Some wondered if opentheword had been hacked. Others asked what language it was. A couple wanted an English translation of the article.
All that happened is I took a story on the website and translated it into Turkish using Google translator:
- English: Former Muslim imam says the Qur’an is the reason he converted to Christianity
- Turkish translation: Eski Müslüman imam, Hıristiyanlığa dönüşmesinin sebebi Kuran olduğunu söylüyor
Here is the back story: A couple of days ago, I received an article in French that I needed to read in English. So I threw it in to Google translator to see what it was about.
I have used the service in the past and it provided a basic, but rough rendering of the article. But this time when it spit out the English translation, I was stunned by how well it read.
The English was clear and readable. The grammar was correct and I had no trouble understanding what it was saying.
I was so surprised, I sent off the translation to a person fluent in French and English to have him take a look. He was similarly shocked by how well it translated.
After a bit of research, we discovered that recently there has been a major improvement in how Google translates nine languages –English, French, Spanish, Turkish, Japanese, German, Portugues, Korean and Chinese.
These nine languages encompass over 35% of the translation requests that google receives through its service.
It is using a new process called Neural Machine Translation (NMT) that basically looks at the whole sentence when translating, and not word by word, as it did before.
According to Google, this one improvement matches all the improvements combined that have taken place in Google translator over the last ten years.
If what I saw in the French to English translation is any indication, they are right.
And I am not the only one to notice.
In an article in the New York Times published last week, Gideon Lewis-Kraus also spoke about the big improvements at Google translate.
He pointed to an article by Jun Rekimoto from the University of Tokyo. Rekimoto specializes in human computer interaction.
Late one Friday night in early November, Jun Rekimoto, a distinguished professor of human-computer interaction at the University of Tokyo, was online preparing for a lecture when he began to notice some peculiar posts rolling in on social media. Apparently Google Translate, the company’s popular machine-translation service, had suddenly and almost immeasurably improved. Rekimoto visited Translate himself and began to experiment with it. He was astonished. He had to go to sleep, but Translate refused to relax its grip on his imagination.
The next day, Rekimoto wrote a Japanese translation of a paragraph from Ernest Hemingway’s book “The Snows of Kilimanjaro.” He passed it through Google translator for an English rendering. He then set this version side by side with the original English and asked people to pick out which of the two paragraphs was Google’s version.
There was only one small error — an almost unnoticeable missing article — in the Google translation. Though different, Google’s version was equally readable and most unfamiliar with the book would not have known which was the Google version.
So after reading this I decided to use Google to translate an article on my site into one of the nine languages that use NMT.
I chose an article that has been on the site’s trending list for months. It is story of how an Imam from India became a Christian after reading the Muslim Qur’an. I know the article has ended up in Muslim forums and is discussed by Muslims.
I translated it into Turkish, because of the nine language listed using NMT it was the only one that came from a prominently Muslim country. With a population of 75 million, only .2% (120,000) are Christian. They attend just 236 churches.
Of course, when I published it, I made a mistake and sent the article out on my English-speaking subscriber list — my apologies.
This is a test case. Depending on how well it does — and provided I can figure out how to stop translated articles from going out on my feed — other translations may start showing up on the site.
You will notice, the Turkish version is on opentheword’s trending list as one of the most read articles over the last couple of days.
Google plans to incorporate NMT in the remaining 94 languages it translates, but provided no time-table.
I have always been interested in providing different language translations on the site, and when WordPress provided a Google translate button a few weeks back, I immediately added it not knowing of the significant improvements underway in Google translate.
PS, if anyone knows Turkish, I would appreciate some feed back on the article.
- Found in translation: More accurate, fluent sentences in Google Translate: Google
- The Great A.I. Awakening: New York Times