The Growing Popularity of Neural Machine Translation

The world is getting smaller, and tools like social media have made it possible to connect with people all over the world. However, language barriers continue to present obstacles. Human translation is the gold standard for communication between cultures, but it isn’t practical for every interaction. Fortunately, advances in machine translation and neural machine translation offer effective solutions for certain situations. 

A Brief History of Machine Translation 

Before the mid-20th century, the only option for translation was to engage the services of a professional human linguist. The concept of machine-based translation was introduced by Warren Weaver in 1949. By 1951, MIT researcher Yehosha Bar-Hillel began the long process of teaching machines to match human skills in this area. 

Early results were not promising. In fact, a 1967 test showed such poor results that the National Academy of Sciences concluded machine translation could never match the level of quality demonstrated by human translators. There was even a suggestion that funding for related research be withdrawn altogether. Nonetheless, the work continued, and machine translation made slow but steady improvement. By the early 21st century, machine translation had a small presence on the world wide web. 

Progress continued, and industry leaders began integrating machine translation technology into new products. Examples include text translation service for Japanese mobile phone users in 2008 and built-in speech-to-speech mobile phone translation for English, Japanese, and Chinese speakers in 2009. In 2012, Google Translate boasted that its service translated enough text in a single day to fill one million books. Though machine translation wasn’t perfect, it offered a valuable service for millions of users. 

The Rise of Neural Machine Translation 

Machine translation’s popularity continues to grow as the technology grows more sophisticated. A new generation of automated translation technology has created a host of new applications for the service. This technology, known as neural machine translation (NMT), outperforms traditional machine translation on multiple levels. 

Research shows that post-translation editing is 25 percent lower when NMT is used versus standard machine translation. While machine translation essentially translates using a word-for-word substitution, NMT translates entire sentences. The results tend to be more fluent, particularly with languages that rely on long-distance relationships between words. German is a good example of such a language. 

Despite the improved performance of NMT, machine translation cannot substitute for human translation 100 percent of the time. NMT focuses on sentences, not full documents, which means that results are still not perfect. In addition, the system relies on machine learning, which takes an extraordinary amount of time and data-input. These resources aren’t available for all applications. 

Researchers continue to attack the outstanding issues with NMT in an effort to further improve results. Leading technology companies such as Facebook and Amazon are developing their own solutions in hopes of using the technology in products and services for their consumer base. Google and Microsoft have already adopted NMT for their translation services, and the customer feedback has been positive. 

For situations that require high-quality translation, there is no substitute for a human translator. People are still the only option for interpreting context, nuance, humor, irony, and other strictly human language quirks. However, traditional machine learning is still a useful tool for circumstances in which basic, low-cost translation is sufficient. Neural Machine Translation offers an opportunity for higher-cost and better-quality results. If this technology continues on its current trajectory, it may eventually match the skill of its human counterparts. 

Learn more about options for translation and narration services at Dynamic Language. 

{{cta(‘8ce8dfba-9ff8-49da-ac9f-7a8339458904′,’justifycenter’)}}