Machine Translation (MT) is the process by which computer software is used to translate a text from one natural language (such as English) to another (such as Spanish). Long disregarded as a buzzword, machine translation has come a long way since it was first introduced in the 1950s, and its evolution has been exponential. Let’s take a few minutes to go over the latest developments in machine translation, and let’s have a look at what the tech giants have been up to lately.
How does Machine Translation work?
Generally, there are three approaches to Machine Translation. The first approach is Rule-based Machine Translation (RbMT). This approach relies on countless algorithms based on the grammar, syntax, and phraseology of a language. The second approach is Statistical Machine Translation (SMT). With lots of parallel texts becoming available, SMT developers learned to pattern-match reference texts to find translations that are statistically most likely to be suitable. Then there is Neural Machine Translation (NMT). NMT uses machine learning technology to teach software how to produce the best result. Many MT providers have switched to NMT, as it is deemed the most promising, most scalable and eventually, most accurate of the three, although some providers utilize a hybrid form as well, combining multiple approaches.
Silicon Valley & Big Tech
For Silicon Valley’s tech giants, machine translation is an indispensable tool to be able to translate the exponentially increasing amounts of online content. Google and Apple have been investing heavily in MT research lately, especially to support their virtual assistants, ‘Google Assistant’ and ‘Siri’ respectively. However, the biggest challenge for Google, Apple, Facebook, etc. is to level the playing field for low-resource languages. Machine learning (ML) models need to be trained, and you need lots of data in order to do so. For rare languages, there simply isn’t enough data to train these machine learning models properly. Consequently, English is used as a ‘pivot language’. In other words, the source language is first translated into English, and then into the target language, because it is easier to train ML models this way. However, Facebook recently boasted a major breakthrough. Facebook’s latest NMT model was designed to avoid English as the intermediary (or pivot) language between source and target languages. Facebook called it the “culmination of years of Facebook AI’s foundational work in machine translation.”
China Joins The Race
Meanwhile, as part of the Chinese government’s three-year action plan to advance the country’s AI technology, including speech recognition and machine translation, Chinese tech giants Alibaba, WeChat and TikTok have been stepping up their MT efforts as well. Alibaba is working on an NMT engine which mimics the human language learning process. They call it “self-paced learning”, and it supposedly vastly improved the accuracy of the engine’s translations. Although these Chinese companies are doing exciting research into MT, Western governments have criticized Chinese MT offerings because they suspect it to be a way for the Chinese government to collect data on users outside of China. Famously, for this reason, the Trump administration said it would ban apps like WeChat and TikTok in the United States, although for now, a judge has blocked the order before it would have gone into effect. Whether Trump’s ban was motivated by security concerns or by a political agenda, it goes to show how important MT is in the fight for technological superiority. Let’s just hope this “technological cold war” ends up benefiting the consumer in the long run.
“As a Language Service Provider, should I be using MT?”
For Language Service Providers, MT is used to augment productivity of their translators, cut costs, save time, and provide post-editing services to clients. In 2016, SDL, one of the largest translation companies in the world, announced it translates 20 times more content with MT than with human teams. So yes, if you want to provide translation services at scale, it’s time to hop on the MT train, if you haven’t already.
Any thoughts, corrections or questions concerning this blog post? Let us know!