It never occurred to the founders of the International Federation of Translation in 1953 that a day would come when machines would threaten the work of translators. Until the first half of the eighties of the 20th century, the system developed by the American company "Weidner Communications" was the only commercially available system for machine translation from English to Arabic.
Although there were two other systems at the time, Systran and Alps, they were not ready for commercialization. Widener was able to market its system to parties in Saudi Arabia, Qatar and the Sultanate of Oman, but these parties were disappointed with the quality of the system after using it.
The 1990s witnessed many attempts to develop automatic translation systems from English to Arabic and from Arabic to English, the most important of which are the Kuwaiti Sakhr Company system developed by the Foundation in Cairo, APPTEK in Washington, and ATA ( ATA) in London, and Simos Arabic (SIMOS) in Paris.
But the qualitative leap in translation from English to Arabic and from there to English came in April 2006 with the launch of Google Statistical MT, which included the Arabic language. Before that, the prevailing machine translation programs were based on Rules-Based grammar. .
On April 4, 2017, Google announced that it had abandoned statistical translation (from and into Arabic), which decomposes entered sentences into independently translated phrases and words (PBMT), and replaced it with Neural Machine Translation (NMT), which is based on deep self-learning that Provides the ability to translate sentences at once without fragmentation.
Many companies today provide high-quality neural machine translation to and from Arabic, including: the translation of the Russian search engine "Yandex Translate" and the translation of "Microsoft Bing Translator".
And last September 14, the latest version of these attempts appeared, as the Jordanian company “Tarjama” launched a translation engine whose address on the web is https://translate.tarjama.com , which the news of the launch described as “the most competitive and advanced machine translation engine ever.”
Why is machine translation important?
Every day the world produces about 2.5 quintillion bytes of data (a quintillion equals a billion billion ie one with 18 zeros to the right of it). Assuming that one word consists of 5 bytes on average, we find that the world of information we are witnessing today produces 500 million billion words per day.
And if we assume that one out of every 100,000 produced words needs to be translated into only one language, and that one translator is able to translate two thousand words per day (about 10,000 bytes), then this means that the world needs 2.5 billion translators, how many translators in the scientist?
It is difficult to give an accurate estimate of this number, but the most acceptable estimates indicate that there are 640,000 translators, who can translate only 1.28 billion words per day. That is why it has become necessary to use smart programs to sort the data produced and classify it according to the levels of importance of its translation and then rely on machine translation to finish the process.
The dilemma of translating new words automatically
According to the Global Language Monitor, the English writer William Shakespeare, during his life from 1564 AD to 1616 AD, created 1,700 new words, and today the English language witnesses the birth of a new word every 98 minutes, that is, about 15 words per day and about 5365 new words annually. .
But the number of these words widely circulated does not exceed a thousand, and intelligent machine translation is not expected to precede humans in translating new words.
Technologies related to machine translation
Many technologies currently available support machine translation, the most important of which are:
Converting spoken Arabic speech to Arabic text (Voice Recognition).
Convert Arabic text to spoken speech (TTS).
Convert spoken speech in English and other languages into text.
Convert text written in English and other languages into spoken speech.
Recognizing Arabic and non-Arabic texts within images (OCR).
These technologies contribute to enhancing simultaneous visual and audio translation, which is what the world needs today, whether in conferences, television programs or video broadcasts over the Internet (via YouTube, Tik Tok, or any other platforms).
The future of machine translation
The results of tests by the American Lionbridge Technologies Foundation indicated that neural machine translation improves annually by between 3% and 7%.
Some studies expect that the level of machine translation will reach the level of human translation by 2030, with the exception of translation of literary and philosophical texts, and those that contain new scientific terms, which are expected to remain in need of modification by humans.
Tags:
DIGITAL LINE