We are overcoming the language barrier - already six years!
Franz Och, Google Translate Team Researcher
Thanks to the development of the Internet, all world knowledge is available to more than two billion users. A short query in the search engine - and you are already on the Internet page, which is stored on the server thousands of kilometers away from you. A couple of clicks - and you read the message from a person who is on the other end of the Earth. But what to do if the information you need is in Hindi, Icelandic or Afrikaans, and you speak only Russian - or vice versa?
In 2001, Google had a service that allowed them to translate from eight languages to English and vice versa. Its core was a state-of-the-art paid machine translation system, but the quality was not up to par, and for a few more years this system was only slightly improved. In 2003, Google engineers decided to improve the quality of translation and increase the number of languages. That's when I started working on this project. At that time, I was still a researcher at the Department of Defense Advanced Research and Development projects (DARPA), where we studied a new approach to machine translation - a self-learning system that was recognized as quite promising and had to significantly improve the quality of translation. I got a call from Google and convinced (and I was very skeptical!) That this approach, based on data analysis, will work in Google.
')
I joined the team, and we began to rework the translation system. We participated in the certification of machine translations conducted by the National Institute of Standards and Technology (USA). This is a kind of grand prix among research institutes and companies, designed to improve the machine translation system. Thanks to the computing power of Google and the ability to process large amounts of information, we have achieved good results. This moment was a turning point: it became apparent how effective the data analysis approach could be.
But at the same time, our system worked too slowly and was impractical: 1000 units of computing equipment “worked” on the translation of 1000 sentences, and in time it took 40 hours. We focused on speed, and a year later, our system could already translate the proposal in less than a second, and the quality was noticeably higher. In early 2006, this technology was used in the English-Chinese and English-Arabic translation and vice versa.
We announced a statistical machine translation on April 28, 2006. For the last six years we have paid the greatest attention to the quality of translation and the increase in the number of languages. Google Translator now works with 64 languages, including those that are poorly represented on the web - for example, Bengali, Basque, Swahili, Yiddish, and even Esperanto .
Today, more than 200 million active users go to the translate.google.com page per month (and even more with other programs that support the service — for example, Chrome, YouTube, mobile apps, etc.). Users also like to use the Translator on the go, because the language barrier is especially relevant when you are traveling: according to our statistics, the use of mobile traffic has grown more than 4 times from year to year. Translator's audience is truly global - 92% of mobile translator traffic comes from outside the United States.
The volume of text that we translate every day is about one million books - or almost the entire volume of professional translations in a year. From these figures it is clear that most of the translations on our planet today are being done by Google Translator (perhaps, in the galaxy, we could have been overtaken by the “ Babylonian fish ” of Douglas Adams). Of course, nothing can compete with professional translators in the transfer of nuances and accuracy in detail. We are sure that machine translation will help more people to participate in the global dialogue in their native language. In this case, the role of professional translators will become even more significant.
Today we dream of a future in which everyone can receive any information and share it with others, regardless of what language it is presented in. We are already translating in real-time web pages in the Chrome browser, texts on photos taken with a mobile phone, captions on YouTube and even oral speech using smartphones. We want to break the language barrier between people - let's see what awaits us in the next six years!
Thanks to the development of the Internet, all world knowledge is available to more than two billion users. A short query in the search engine - and you are already on the Internet page, which is stored on the server thousands of kilometers away from you. A couple of clicks - and you read the message from a person who is on the other end of the Earth. But what to do if the information you need is in Hindi, Icelandic or Afrikaans, and you speak only Russian - or vice versa?
In 2001, Google had a service that allowed them to translate from eight languages to English and vice versa. Its core was a state-of-the-art paid machine translation system, but the quality was not up to par, and for a few more years this system was only slightly improved. In 2003, Google engineers decided to improve the quality of translation and increase the number of languages. That's when I started working on this project. At that time, I was still a researcher at the Department of Defense Advanced Research and Development projects (DARPA), where we studied a new approach to machine translation - a self-learning system that was recognized as quite promising and had to significantly improve the quality of translation. I got a call from Google and convinced (and I was very skeptical!) That this approach, based on data analysis, will work in Google.
')
I joined the team, and we began to rework the translation system. We participated in the certification of machine translations conducted by the National Institute of Standards and Technology (USA). This is a kind of grand prix among research institutes and companies, designed to improve the machine translation system. Thanks to the computing power of Google and the ability to process large amounts of information, we have achieved good results. This moment was a turning point: it became apparent how effective the data analysis approach could be.
But at the same time, our system worked too slowly and was impractical: 1000 units of computing equipment “worked” on the translation of 1000 sentences, and in time it took 40 hours. We focused on speed, and a year later, our system could already translate the proposal in less than a second, and the quality was noticeably higher. In early 2006, this technology was used in the English-Chinese and English-Arabic translation and vice versa.
We announced a statistical machine translation on April 28, 2006. For the last six years we have paid the greatest attention to the quality of translation and the increase in the number of languages. Google Translator now works with 64 languages, including those that are poorly represented on the web - for example, Bengali, Basque, Swahili, Yiddish, and even Esperanto .
Today, more than 200 million active users go to the translate.google.com page per month (and even more with other programs that support the service — for example, Chrome, YouTube, mobile apps, etc.). Users also like to use the Translator on the go, because the language barrier is especially relevant when you are traveling: according to our statistics, the use of mobile traffic has grown more than 4 times from year to year. Translator's audience is truly global - 92% of mobile translator traffic comes from outside the United States.
The volume of text that we translate every day is about one million books - or almost the entire volume of professional translations in a year. From these figures it is clear that most of the translations on our planet today are being done by Google Translator (perhaps, in the galaxy, we could have been overtaken by the “ Babylonian fish ” of Douglas Adams). Of course, nothing can compete with professional translators in the transfer of nuances and accuracy in detail. We are sure that machine translation will help more people to participate in the global dialogue in their native language. In this case, the role of professional translators will become even more significant.
Today we dream of a future in which everyone can receive any information and share it with others, regardless of what language it is presented in. We are already translating in real-time web pages in the Chrome browser, texts on photos taken with a mobile phone, captions on YouTube and even oral speech using smartphones. We want to break the language barrier between people - let's see what awaits us in the next six years!
Source: https://habr.com/ru/post/142914/
All Articles