How does Google Translate work?

Whether you’re a student or just someone you need to translate a word or phrase is likely that you may use translation service from Google, this service allows you to translate from different languages and offers many useful services, which made us dispense with many translation software after proved It is better than it is in many cases but never thought of asking yourself how Google Translate and those similar services work ?

With your use of the site and its service, you may have noticed that Google Translate provides instant translation and includes the translation of sentences and words. To convey the meaning, we certainly are not here today to introduce the service, as I mentioned earlier in the list of browsing history is not free from this site, but today I will try to answer your question and explained how the service works.

At first glance, you might think that there is a place where thousands of employees from all over the world come together and speak all languages. Well, if we talk about your imagination in two separate parts, yes, there are thousands of employees from all over the world working at Google, but certainly not their job is to enter data manually, you see how the Google translation service works to translate every word Enter them into it? Continue with the next lines to know.

About Google’s translation service and its history

Google Translate is a free service developed by Google, you can translate texts, conversations, photos, locations instantly, between 103 supported languages in addition to the fourteen languages are developed. The service was first launched in April 2006, ie April of the previous year was its 10th anniversary. The site is visited and used by over 200 million visitors a day. The service initially started as Statistical Machine Translation, until in November 2016 it was switched to Neural Machine Translation, which we will explain as much as possible in the following lines.

Statistical Machine Translation

The technique began in the early 2000s when translation machines develop complex algorithms for the most accurate translations, but of course, they do not amount to translation such as human beings.

Like other translation sites, the site works through statistics, by inserting millions of human-translated documents and books, as well as United Nations documents, which are translated into the six major languages, thus providing language data. Introducing many translation options for sentences, which have already been introduced by humans, and then selects the closest, most commonly used, and most common translations.

This is the difficulty. It is not as we have mentioned a huge number of human beings sitting to enter words and their meanings word by word, but the difficulty is to find this huge linguistic content. For example, you will find that the content written in English is more than any other language, which makes you may find that the translation to and from English gives better results than any other translation between two other languages.

English has been used as an intermediary between any two languages, because if you want to translate a word from Arabic to French, for example, Arabic will first be translated into English, then English to French, and of course all this will happen in the least part of the second. This means that when translating between two languages ​​that do not have English, the translation machine completes two consecutive operations, showing the greatness behind the construction of this genius system of course.

In addition, there are some languages ​​where there are more than two operations. For example, if you want to translate a Catalan word into Chinese, the site will first translate the Catalan word into Spanish, which is the origin of the language. It then translates the Spanish word into English, and then from English to Chinese, which is also done in a fraction of a second.

Prior to 2007, SYSTRAN was translated into languages ​​such as Arabic, Chinese, and Russian, an engine that is still being used by many online translation services such as Babel Fish, until the transition from October 2007 to the system. The former is Statistical Machine Translation.

SYSTRAN, founded by Dr. Peter Toma in 1968, is one of the oldest translation companies in the US defense department, providing technology to Yahoo and several other sites.

There are some words that are so complex can not be used for example in articles intended for a simple audience, but they can be used in poetry, etc. Google does not know from the audience to which the translated sentences are, and they are not specialized translations, so they can give you simple translations when you don’t need them, and very complicated translations when you don’t need them.

Google Statistical Translation does not have feelings, as we mentioned earlier, this translation system from Google is based on statistics, and the best translation for him is the most used by translators, regardless of where and where they use these words, they do not give you a sentence The most influential in the reader, or the most attractive sentences and words, even if the translation is accurate, it is still devoid of living emotions of course, and this appears in the voice of the reader, which you may sometimes use in the pronunciation of words to learn how to pronounce the correct, it speaks without feelings, only electronic sound Free from anything related to emotions Humanity.

The grammar of languages is another important reason for the weakness of Google’s translation service and other translations. – Free of gender discrimination in terms of grammar – while translating other languages, making sentences free of rules when translating languages ​​called romantic languages ​​because of the grammatical distinction between masculine and feminine.

Neural Translation Machine

At the end of 2016, specifically in November, Google’s translation system moved from statistical translation to neural translation, a translation style that uses deep learning techniques. This system was developed by Google itself.

Google’s Neural Translation Machine uses a huge artificial intelligence neural network that is capable of deep learning through the use of millions of examples. Human Translation Rules.

This system came closer to the human level of translation, as it was able to overcome the obstacles faced by the statistical system when translating the complete sentences between Chinese and English, through translation by making use of bilingual humans to teach the system. In other words, Google simply increased the sources of words and sentences entered into the system.

The company has launched what is known as the translation of Zero-Shot translation system from Google multilingual, which avoids the translation of conjugal terms in the same language by introducing phrases into the system have meaning independent of the language.

For example, in their report, they describe how Japanese-English and Korean-English translation can be used together to teach the multilingual system, and then request the translation of pairs of words that have never been presented before, which is surprisingly Korean-Japanese translation. , The system was able to produce sentences with real meaning and by translating between only two languages ​​without reference to intermediate English between them.

It is not as simple as we expected, there are hundreds of highly intelligent minds working trying to reach the best results through this translation service from Google, which may be seen by some as simple in design and shape, or complicated in the way they think they work and that We mentioned it at the beginning of the article, people trapped on an isolated island, who enter thousands of words a day into service.

This has reached the highest levels of technology currently available through the use of artificial intelligence, and the ability to develop the machine to be able to teach itself by absorbing as much of the information provided to them, and provide simple to the user.