Machine Learning Makes Inroads Into The Intricate Art of Translation

Language and writing expert Reuven Koret discussed in detail the state of influence and use of artificial intelligence in translation for the online publication readwrite. Koret points out that the use of machine translation tools based on AI in all aspects of the translation process is becoming widespread. This is not solely reserved for proprietary ML translation tools from Google, Microsoft,  Facebook, and Amazon are in daily use, but detailed professional tools from companies like SDL.

Still, many professional translators and agencies like William Mamane, Head of Digital Marketing at Tomedes, a professional language services agency are still skeptics about the use of AI in translation. But even those skeptics like Mamane admit that machine translation has made serious advances, and as he points out, “there still is a place for AI and Machine Translation in the translation services value chain.”

To explain the challenge of machine translation, Koret notes that “at a basic level, MT uses algorithms to substitute words in one language for those in another. That proves insufficient to translate successfully. Understanding of whole phrases is necessary for both source and target languages. We can understand MT as decoding the source language and recording its meaning in the target language.”

Resolving this challenge is a very complex process and currently, the most developed processes are using “statistics to choose the best translation for a given phrase,” or “structured rules to select the most likely meaning.” These approaches still require the engagement of editors and proofreaders, but “that supervisory, editorial, or auditing role is less demanding and less time-consuming than translation.”

These methods are the ones on which most web translation apps like Google Translate are based on. As is noted, Google processed translations that would fill one million books per day. 

Currently, though, even bigger strides in using AI in the translation process are accomplished with the use of neural machine translation (NMT), Using deep learning when translating, “looks at full sentences, not only just individual words.” At the same time, NMT requires “a  fraction of the memory needed by statistical methods,” meaning that at the same time it works much faster.

The use of NMT was first researched only in 2014, but the rapid advances in the last five years have made it possible for the development of the bidirectional recurrent neural network or RNN. “These networks combine an encoder which formulated a source sentence for a second RNN, called a decoder. A decoder predicts the words that should appear in the target language.” Google is no using this approach in the NMT to drive Google Translate. Also, Microsoft uses RNN in Microsoft Translator and Skype Translator.

As Koret concludes, NMTs can assist in translating while skilled linguists can finish and polish the translation output. Future translators will be more often working with artificial intelligence rather than against it.”