GOOGLE STILL LOST IN TRANSLATION

Tech giant unveils a new system for Arabic translation that it says has more natural language fluency and accuracy

Tech giant claims it has improved its Arabic. But a translator says it’s still awful,

ABU DHABI // Translation between Arabic and English is difficult enough – but anyone who has tried to use automated translation realises that this is even more problematic. So Google’s announcement yesterday that it has greatly improved its Arabic translation will come as a relief to many.

Thousands of Arabic speakers and millions of their phrases have been used in new technology that is capable of translating whole sentences, rather than interpreting text piece by piece and learning over time from the input.

Google’s “neural machine” translation system creates more natural translations, leading to increased fluency and accuracy.

Since last November, the US technology giant has launched its new system for 15 other languages.

The previous translation system, used for the first decade of Google Translate, relied on phrase-based technology.

The old and new systems rely on massive amounts of translation information, but the main difference is in how they apply that data, said Najeeb Jarrar, Google’s product marketing manager for the Middle East and North Africa.

“Neural doesn’t really learn languages as such, but rather it teaches itself how to generate results that match the parallel translations we’ve given it to learn from,” he said. Maya Mazloum, who teaches Arabic at the Eton Institute, said she has noticed a marked improvement in Google’s translation of an English article into Arabic.

“When I first saw the translation I said ‘wow’,” said Ms Mazloum, who has taught Arabic for nine years.

“I felt that it didn’t need the usual amendments. The order of the words was better and the grammatical endings were matching the gender of the nouns.”

Although the new system will help her students, who regularly use Google Translate, translations still need guid- ance from native speakers who can check and amend them, she said.

Despite its limitations – especially for colloquial Arabic – the technology is a welcome aid in the classroom, said Ms Mazloum. But Hani El Imam, owner of Smart Translation, a UAE company, and an Arabic- English translator for 12 years, said the new system has degraded the quality of what he described as an already unreliable service.

Although the new system has improved the grammar, Mr El Imam said he was finding an increasing number of incorrect interpretations of certain words. “No matter how far you advance the technology, you will never be able to replace human translation, especially with a rich language such as Arabic,” said Mr El Imam, who wrote his thesis for his master’s degree on the incorrect translation of metaphors in the Quran.

He said translation problems arose from linguistic and cultural differences between languages that challenge even the best translators.

“If the leading translators are still translating metaphors in the Quran incorrectly, are you telling me that Google will be able to translate them in the regular language?” Mr El Imam asked. “I doubt it.”

But Mr Jarrar said Google’s translation system was not designed to supplant the incredible nuances of professional translation. “This system can’t translate something sophisticated like a novel, with all its contextual and cultural subtleties,” he said. “But we’ve heard some translators use it for a quick first pass that they can then refine.” The more Google’s system is used and as it learns from suggestions from the Google Translate community, the technology would continue to improve, said Mr Jarrar.

“This is only the beginning and we expect the quality of the translation to get better,” he said.