The Google Translate app’s new neural network technology is taking a novel approach to translation, with the system devising its own language that uses use contextual techniques to perform interpretation. This artificial language has proven so effective it can even perform translations between two tongues it wasn’t specifically trained to support.
Google researchers say their system can identify commonalities between sentences in different languages that have the same meaning, as reported by New Scientist. This approach is made possible by the neural network’s capability to understand complete sentences.
The self-learning end-to-end learning system built into Google’s neural machine translation technology creates better and more natural translations since it comprehends each sentence. This leads to smoother and more readable translations of paragraphs and articles.
This new system is coming to eight of the 103 languages supported by the Google Translate app. They are French, German, Spanish, Portuguese, Chinese, Japanese, Korean and Turkish. Google says those languages represent 35 percent of all translations performed on the service.
This contextual approach also enables greater accuracy than before, cutting down translation errors by 55 to 85 percent, CNET reports.
Google’s breakthrough reflects advances that have vastly improved translation technologies during the past few decades. Machine translation has evolved from literal word-to-word interpretations to sophisticated techniques, including semantic translation techniques that convey meaning, faithful translation systems that convey the author’s intention and contextual translation approaches that take context into account for better comprehension.
Pratik Dhebri is senior director of product management at Veritone and is truly democratizing Artificial Intelligence (AI) by orchestrating best-of-breed cognitive engines. He works with AI developers and cognitive engine providers to architect and monetize algorithms and applications. Learn more about our platform and join the Veritone developer ecosystem today.