Neural machine translation: Bridging the gap between human and machine
Many companies are interested in using machine translation for their businesses. It offers a number of advantages over traditional human translation, particularly when dealing with very large volumes of text.
For one, it is much, much faster. A human translator can translate a maximum of 2000 - 3000 words per day, but machine translation can do the work of hundreds of human translators. Of course, this also means that machine translation is much less expensive than human translation. The cost savings can be enormous when working with large projects.
However, the quality of machine translation is considerably lower than with human translation. The translations are often far too literal, the grammar might sound strange, and the language used is often unidiomatic. This is because of the way that traditional machine translation works - by translating set phrases or sequences of words.
Statistical Machine Translation (SMT) has been the dominant translation paradigm for decades. Practical implementations of SMT are generally phrase-based systems (PBMT) which translate sequences of words or phrases where the lengths may differ
— Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation, 2016.
Neural machine translation is working to change that by creating a whole new way of processing the texts that are being translated.
What is neural machine translation
Neural machine translation is an advanced new method of machine translation. Traditional machine translation analyzes a sentence and then attempts to replace the words in the sentence with equivalents in the second language. In a series of processes, it replaces the words, changes the word order so that it is appropriate to the target language, and it tries to make sure that short word groupings are fluent sounding. However, the results are usually less than stellar. Neural machine translation is different.
Unlike the traditional phrase-based translation system which consists of many small sub-components that are tuned separately, neural machine translation attempts to build and train a single, large neural network that reads a sentence and outputs a correct translation.
— Neural Machine Translation by Jointly Learning to Align and Translate, 2014.
Neural machine translation uses a totally new method of processing data, and this method is revolutionary. Neural machine translation programs use a large neural network that allows the translation program to learn from its mistakes and really “understand” the languages it is working with. The exact process is difficult to explain in detail, because many of the steps are hidden within the programming, but the end result is customized machine translation that is far better than what standard machine translation can produce. The results are far more natural and fluent sounding, and require significantly less work to make them ready to publish.
The strength of NMT lies in its ability to learn directly, in an end-to-end fashion, the mapping from input text to associated output text.
— Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation, 2016.
What does that mean for translation buyers?
Neural machine translation can significantly improve productivity in the localization process, even compared to traditional machine translation. For example, it reduces the need for post-editing by 25%. That means that the overall process is considerably faster than with regular machine translation, because less time is needed for editing. This also means significantly lower translation costs - because, as we all know, time is money.
What can it be used for?
Neural machine translation is a great option for companies who need to localize large volumes of text efficiently and cost-effectively. In this day and age where many companies produce large quantities of written content in the form of blogs, informational articles, social media channels, videos, and FAQs, traditional translation just isn’t fast enough to keep pace with the volume of content being produced.
The results of traditional machine translation leave a lot to be desired in the quality department. However, neural machine translation is a state-of-the-art option that can provide, fast, efficient localization with significantly better quality than what traditional machine translation can deliver. The program can even be “trained” using existing translations, so that it can achieve a tone that is close to that which has already been used. This makes it perfect for translating large volumes of online content with a good level of quality.
