Most people have used neural machine translation (or neural MT), but few need to understand what the technology entails.
However, companies using translation technology to maximise the efficiency of their translation process need to know why neural MT is the industry-standard – and why it is more capable than the other tools available.
Neural machine translation is the key technology behind the leading corporate and consumer translation tools, including Google Translate, Microsoft Translator, Unbabel and our own MT plus solution.
“Neural Machine Translation is a machine translation approach that applies a large artificial neural network toward predicting the likelihood of a sequence of words, often in the form of whole sentences. [Neural machine translation] systems are quickly moving to the forefront of machine translation, recently outcompeting traditional forms of translation systems.” – What is Neural Machine Translation?, DeepAI.org.
To explain the mechanics of how neural machine translation works, it helps to start by defining its two key components:
- Machine translation
- Neural networks
We’ve recently published an article answering this question, in which we define it as follows: “machine translation (MT) is a computational translation technology that uses programmes to translate text or speech from one language into another.”
There are three main types of machine translation commonly used today.
Rule-based machine translation
Rule-based machine translation uses extensive sets of linguistic rules developed by language experts to translate source content into the target language.
Statistical machine translation
This analyses existing translations performed by professional translators, to determine the most likely translation for the source content.
Neural machine translation
Neural machine translation uses neural networks to “learn” from existing translations and its own previous translations to constantly improve its output.
Neural machine translation has become the leading standard of translation technology, using artificial intelligence (AI) and machine learning to process huge volumes of data to make decisions and improve results without human input.
If we delve deeper into the AI and machine learning powering this type of translation, neural networks are the key technology.
A neural network is a machine learning technology designed to replicate the neural behaviour of the human brain. In simple terms, what neural networks do is mimic the pattern recognition that drives human learning through repeated exposure to situations and a variety of outcomes.
IBM has an excellent resource on neural networks that explains their technical functionality in more detail:
“Artificial neural networks (ANNs) are comprised of node layers, containing an input layer, one or more hidden layers, and an output layer. Each node, or artificial neuron, connects to another and has an associated weight and threshold. If the output of any individual node is above the specified threshold value, that node is activated, sending data to the next layer of the network. Otherwise, no data is passed along to the next layer of the network.” – What are neural networks? IBM.com.
Neural networks require training data to identify patterns that meet specified thresholds and improve their ability to perform tasks successfully. Fully developed neural networks are capable of training themselves without human input to continue improving their performance – as is the case with neural machine translation systems.
While rule-based MT moves from word to word and statistical MT is capable of translating complete phrases, neural machine translation has the ability to analyse the contextual meaning of full paragraphs.
By piecing these paragraphs together, neural MT can understand the meaning of your entire content and use this to inform its translation decisions. In essence, this mimics the cognitive process professional translators use to translate content.
While neural networks cannot really match the thinking power of the human brain, neural MT is a significant step up from previous translation technologies in that regard. It achieves this through the following ways:
- Big data: Neural MT uses far more data than other solutions to make smarter, data-driven decisions.
- Contextual understanding: Neural networks are capable of understanding the contextual meaning of content, not only the literal word-for-word meaning.
- Greater accuracy: Powered by machine learning, neural MT is more accurate than other forms of machine translation.
- Fast-learning: Neural networks learn from training programmes without manual data management.
- Accessible: You can train and implement neural networks into your own software without needing to know how they work.
- Constant improvement: Neural MT trains itself without human input and constantly improves its results.
- Speed: Like all forms of machine translation, neural MT produces almost instant results, but with greater accuracy.
- Cost-effectiveness: The open-source nature of neural MT and the improved accuracy over alternatives make it the most cost-effective machine translation solution.
Translation technology isn’t about achieving perfection, but increasing the efficiency of translation processes. After a series of breakthroughs in AI technology over the past couple of decades, we’ve reached a point of diminishing returns, but every incremental improvement reduces the expense of managing translation projects.
This is why it is so important to use the right technology – because every gain can save companies money or allow them to achieve more with the budget that is available.
If you want to talk neural machine translation and how we use it as part of our translation technology and AI translation solutions, our MT consultants can help. Please submit any relevant MT project requests you may have via this contact form.