AI translation App-Main Art-1582365644046


Time was when you took along a hefty edition titled ‘100 Common Phrases in Italian’ (or some other language depending on your destination) on your foreign holiday to help you communicate. These days, all it takes is an app. Once impossible to imagine, technology that allows people to communicate instantly with each other in different languages is now a reality.

An earpiece that allows two people who speak different languages to communicate and understand each other? Yes, thanks to Waverly Labs, the Brooklyn-based firm. Converting foreign language text into the language you read? Yes. There are a host of options including apps such as iTranslate and Translate Voice. Want to know what that signpost means on your travels? Google Translate app’s camera function will help you. Just point and know.

Experts say AI powered translation will change the way we live. Schools could gain access to education that would otherwise be foreign to them and politicians could exercise their diplomatic flair with a host of nations. According to Dr Saihong Li, professor of translation studies at the University of Stirling, UK, translation could become so advanced that people won’t even know translation is taking place. “Translation technology really plays a crucial role in politics, travel and in everyday life.”

At the 7th Abu Dhabi Translation Conference held in the UAE last November, one of the workshops was themed: ‘Can a Robot Replace the Translator?’

The answer is not a straightforward yes or no.

According to Dr Dipankar Kundu, CEO and Founder of the Dubai-based Dar Al Marjan Translation Services, Neural Machine Translations (NMTs), an AI-powered software that works as a language translator, can definitely help but can never supersede the skills of human translators. “The neural network is a system that can be trained to recognise patterns in data, thereby transforming input data in one language into a desired output in another language,” explains Dr Kundu, who has a PhD in Arabic literature. ”But before any translation can happen, the human engineers need to decide the architecture of the network. The human engineers are technology masters but are not masters in languages and translation. Also, the guesswork of input data to output data is not only difficult but leads to erroneous translations.”

NAT Dr Dipankar Kundu-1582365648340
Dr Dipankar Kundu

And herein lies a world of hits and misses. The world of translation is vast and infinitely complex. On the translation spectrum, you have everyday phrases such as ‘How are you?’ ‘Where is the airport?’ at one end to the highly complex medical literature, legalese, literary allusions, idioms and context, poetic nuances and old manuscripts at the other. Plus, there is the matter of dialects.

Says Dr Kundu, “For example, Arabic has more than 30 dialects. In French, there are 28 dialects. Similarly, every language has a number of dialects. The translator needs to know the target readers and have knowledge about their culture. NMT cannot do this.”

Another area: literature. “Literature is an ornamental form of writing characterised by extensive use of idioms and figures of speech,” says Dr Kundu. “The literal meaning of an idiom is definite when it is discrete but gives different connotations depending on its contextual use. A figure of speech needs to be interpreted according to the story line. NMTs will never have the ability to understand the context,” he says. Plus, NMTs will fail to make value judgments, he adds.

Similarly, the translations of Old Manuscripts is a challenge that is fit perhaps only for the human brain, says Dr Kundu. “The study of old manuscripts have made available to human kind a lot of beneficial information about the past be it in science, technology, history, medicine, civilisation, architecture.... Since, all the data collected by AI-powered software is from the web, NMT cannot understand these documents due to the works being handwritten documents, primitive script, unfamiliar grammar, syntax, semantics and vocabulary and not enough data available in web to support translation of such a language and text of such nature,” he says.

Even in the field of science, “Every day there are new inventions and discoveries. Translation of new scientific data by NMTs will fail as NMT is an AI powered software and can only identify and translate data that has been fed in,” says Dr Kundu.

Matteo Ippoliti, Founder and General Manager of LangPros Dubai believes that when it comes to language learning, AI is forging ahead. “AI will play a major role when it comes to learning new languages, communicating in a foreign country, translating documents or interpreting during events,” he says. The tourism and travel industry already has apps using Machine Translation (MT) and AI helping travellers and tourists. “But at the moment, these apps would work mainly for basic conversation and not for complex communication situations.”

NAT Matteo Ippoliti-1582365650579
Matteo Ippoliti

In the past few years, there is been unprecedented progress in the area of machine translation especially with Neural Machine Translation, says Ippoliti, which is employed by tech giants such as Google or Microsoft. “Machines are actually able to form neural networks and learn and improve the quality of translation, in many fields. Despite this progress, the human component is still necessary.”

Mastering a foreign language for translation and interpreting means knowing much more than just the language, says Ippoliti. “Translators and interpreters should have deep knowledge of the culture and the history of the people who speak a particular language. They should understand nuances, references and other important elements of human communication that machines still cannot grasp,” he says.

A human brain’s understanding and perception of the world and its reality is far superior to a machine’s grasp of this domain, according to Ippoliti. “A machine, at least to this day, still cannot reach [there]. Figurative language, irony, sarcasm, poetry, are still out of a machine’s capability as they cannot feel what humans feel when using language. This does not mean we will never get to a point where machines will be able to feel, but it will take time and it will be a very different world, a new era for mankind,” he says.

“At the moment, we are still far away from creating a machine that can behave and feel the reality like a human. Since a language is used to express this unique relationship between humans and the world, sometimes machine translation may lead to major mistakes and to embarrassing or funny results.”

The human component, according to Ippoliti, is still indispensable to eliminate mistakes, adjust meaning and style and optimise the result for a high-quality translation.

What are the most challenging languages to translate? “Those spoken by a limited number of people, such as regional languages or “dialects” where there are only a few translators available. For instance, aboriginal languages in Australia or certain tribal languages in Africa,” says Ippoliti.

The Arabic language and translation

“Arabic is not written fully vowelled and thus a human who knows grammar can read a word with proper declination, understanding the case,” says Dr Kundu. Correct reading gives the correct meaning. Also, in Arabic morphology, a word, though it looks the same, needs to be read with different declinations according to different cases in different sentences and this applies to each and every word in Arabic. These factors make NMTs not the right instrument to be used in translation.”

When it comes to Arabic, the situation is relatively complex, according to Ippoliti.”Although there are many varieties of Arabic (according to some, approximately 30, including Classical Arabic) due to the fact that it is spoken in several countries, the variety that is generally accepted as standard is the Modern Standard Arabic, which is the one used and mastered by most professional translators and interpreters.”

The worth of human translation to build cultural bridges and meld the various chapters of history for a new future is at the heart of the UAE’s ‘Translation Challenge’. Launched by His Highness Shaikh Mohammed bin Rashid Al Maktoum, Vice President, Prime Minister and Ruler of Dubai, in September 2017, it is a key initiative that aims to create advanced educational content in science and mathematics for Arab youth.

A year after its launch, the challenge has translated 5,000 educational videos covering various scientific topics, such as physics, chemistry, biology, mathematics and general sciences, into Arabic, totalling some 11,207,000 words and 50,000 minutes, at a rate of 500 videos per month.

The challenge attracted over 51,000 volunteers from around the world.