Machine Translation 2018: Google vs. Bing vs. Yandex
Hello everyone, my name is Alex and I’m the Founder of Translation Cloud. We’ve just posted a video to YouTube on the state of Machine Translation in 2018.
Machine translation has traditionally been a difficult task to accomplish. But with the constant improvements in computer technology, the quality of machine translation has been slowly getting better over the years. And just as computers first learned to play checkers before the mastered chess, perhaps the next evolution of human-quality machine translation is just around the corner.
The early days of machine translation.
When machine translation began, computers were pretty rudimentary and were not able to provide complex translations. Translation was mostly rule based—Babelfish, by the search engine AltaVista, pioneered this method which basically used two dictionaries. The program applied a set of linguistic rules to try to extract the meaning of the words and (hopefully) provide a meaningful translation. But, as we know, all languages have exceptions, and syntaxes and rules are not a good indicator that a machine can understand the meaning of the text and produce a good result.
In 2006, Google had long supplanted AltaVista as the leading search engine. They wanted to develop their own machine translator because Google is a multinational search engine, and they deal with all kinds of content, not just English-language content. Additionally, they wanted to be able to be sure that their index is clean, and that SEO specialists were not gaming the system, they developed their own machine translator to make sure they could understand what is and is not spam. Google Translate began as a rule-based translator, but through innovation transformed it using a statistical approach. This method allowed users to provide corrections, which built a database that allowed the system to learn from mistakes and provide better and better results.
Besides Google, there are a few other major players and competitors in the online machine translation world, including Yandex, the Russian search engine; Promt online dictionary, and a handful of others.
Machine translation today.
Today, Google Translate uses a “neural network” for translation which mimics human thought patterns and decision making. The more it translates, the artificial intelligence becomes more and more precise, and its “forecasts” are being carried out with greater precision. In 2017 there was big news that Google Translate had passed the “Turing test,” which in essence meant that under certain conditions it was able to produce a translation that was indistinguishable from a human. People were amazed at how good the quality was.
As the internet expands exponentially, and more and more content is created and posted every day—social media, SEO content, landing pages, articles, blog posts, and so on—the possibility of complete human translation of all the world’s content becomes more and more difficult. There is simply not enough time, money, or resources to be able to commit to the task. However, with the growth of machine translation and the use of application programming interfaces (APIs), it is easier than ever to budget for adequate translation of basic and urgent content.
Comparing machine translation services.
The key players of machine translation today are Yandex, Bing, and Google. Use the table below to compare their services and what they offer, and use it to make a decision for your translation needs.
| Yandex | Bing | ||
|---|---|---|---|
| Free Tier | No | Yes | No |
| Price per Million Characters | $15 | $10 | $20 |
| Neural Network Translation | No | Partial | Partial |
| Number of Languages | 95 | 60 | 100+ |
Google has the best quality of translations, but the most accurate, neural network methods are currently only supported by the most popular languages (Spanish, Russisan, French, Japanese, and so on). Bing is catching up and offers similar translation methods for some languages, but Yandex is still solely a rule-based translation service.
For our own internal projects at Translation Cloud, we have typically stuck with Bing as it allows us to develop projects with a free usage up to a certain limit, the quality is getting better with neural networks, and the price is the lowest of all three providers. Even though it only provides 60 languages currently, when looked at critically, it has been shown that you can reach 80% of the world by offering content in only the 10 most popular languages, so this has typically been more than sufficient.
If you learned something from this video, please give it a like and subscribe to our channel!
