You are likely already familiar with machine translation, thanks to the tech powerhouse Google, and their easy-to-use “Google Translate” feature, which automatically translates text in over 100 languages. Historically, companies have used statistical machine translation, a technology that translates language through a series of phrase-based models. While a great first step, with statistical machine translation the nuances of language are lost and the translated text is often unnatural, awkward or even blatantly incorrect.
As we move into 2018, a new dawn of machine translation is upon us. With the development of neural machine translation, we will see the language industry evolve over time with an increase in demand for language services.
THE HISTORY OF MACHINE TRANSLATION
Machine translation emerged as a subject of academic research in the 1950’s. The Georgetown-IBM experiment in 1954 was one of the first successful projects, during which researchers were able to automatically translate over sixty Russian sentences into English. The practical application was still decades away, but this early experimentation proved the possibility of fully automatic translation. The first functional rule-based machine translation software emerged during the 1970’s, with tech pioneers like Systran paving the way.
Further developments in machine translation led to the adoption of statistical machine translation in the 1980’s and 1990’s, and this system has continued to be our primary machine translation technology until now. In recent years, researchers and technology companies have developed a new system: neural machine translation, which fundamentally approaches translation differently and produces more natural results.
NEURAL MACHINE TRANSLATION: A LEAP FORWARD
Statistical machine translation has advanced dramatically throughout the past few decades, but the technology is flawed. Text, particularly large blocks of text, translated using statistical MT can often sound awkward to a native speaker. Statistical MT translates text word-by-word rather than considering the context of each word within the sentence. As a result, individual phrases within the translation may be correct, but the full text reads awkwardly.
Neural machine translation provides more accurate, natural results by analyzing each word within the context of the complete sentence. Neural machine translation takes into consideration the similarities between words: unlike statistical MT which uses numerical substitution by assigning random numbers to words and performing mathematical equations to find their foreign-language equivalents, neural MT recognizes that many words are closely related and assigns numerical values based on those connections.
TECHNOLOGY COMPANIES IN THE GAME
Several large tech companies including Amazon, Google, Microsoft and Systran have launched neural machine translation systems over the past two years. Google Translate now operates using Google Neural Machine Translation, and Amazon is currently previewing Amazon Translate, which allows users to localize content for international consumers easily and inexpensively.
APPLYING NEURAL MACHINE TRANSLATION IN YOUR BUSINESS
Neural machine translation is a huge leap forward, providing better results for the everyday consumer and increasing the number of languages that can be automated effectively. While most neural MT systems are still in development, technology companies are racing to release systems in 2018. So where does this leave the human translator? The truth is, even in today’s world abuzz with AI and machine learning advancements, the human translator still has an important role to play. Neural machine translation does not learn languages in the same way that humans do and cannot quite understand cultural idioms, human emotions, and linguistic nuances. For legal or medical related content, a human eye will still be critical to ensure proper terminology use and sentence structure, informing the meaning of a particular concept or word. The good news is that NMT’s application does reduce post-editing efforts by 25% when compared with statistical MT, and produces more fluent sounding results, meaning translations will be produced at more cost-effective rates and faster time to market.
As machine translation improves, expect to see an explosion of international content. Companies that recognize the potential within their international markets will now be able to more easily connect with foreign consumers by translating their content into multiple languages. As this cost-cutting technology greatly reduces the amount of time and energy human translators spend on basic content translation, similar budgets can be allocated to providing even more language combinations to second and third tier markets. As a result, professional translators will be able to focus on editing translated documents, adding value where it is most important.
The human translator isn’t going anywhere, but with neural machine translation tools, companies will now be able to allocate resources differently. Fluent, native speakers will catch the subtleties of language missed by translation software, and add value and insight in highly technical translation, creative content or cultural appeals. The result is powerful: localized content that a foreign audience will not only understand, but enjoy, at a reduced cost to the company.