Artificial intelligence (AI) has powered some of the most significant advances in translation technology over the past decade. With breakthroughs in machine learning technologies, such as neural networks, natural language processing and deep learning, there is a lot to be excited about with AI translation.
However, there is also a lot of hype and speculation surrounding the technology.
In this quick guide, we cover the essentials on artificial intelligence translation, dispel a few common myths as well as explain how language professionals use the technology to deliver results faster without losing out on translation quality.
AI translation is the most advanced approach to machine translation, using a range of machine learning technologies to gain a deeper understanding of language. Artificial intelligence has progressed significantly over the past couple of decades, largely thanks to advances in two key areas: computer processing power and machine learning technology.
Machine learning uses intelligent algorithms to analyse vast amounts of data and identify patterns, allowing it to “learn” and replicate automated tasks, including translation.
Within machine learning, the development of a technology called neural networks and methodologies called deep learning have produced the most advanced iteration of AI translation to date: neural machine translation (NMT).
For a more detailed look at neural machine translation, check out following article on our blog: What is neural machine translation?
In this article, we quote the following definition from DeepAI.org:
“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.”
Neural machine translation is the technology currently powering the most advanced AI translation tools in the industry, including Google Translate, Microsoft Translator, Unbabel and our own MT solutions.
Essentially, a neural network is a specific type of machine learning algorithm that replicates the neural behaviour of the human brain. More specifically, it mimics human learning by repeating the same task and learning the sequences of events that achieve the desired outcome most effectively.
On a large scale, these algorithms train themselves to translate between languages, learn from their own mistakes (and successes) and continue to improve by analysing vast amounts of data. Google Translate, for example, analyses billions of translations across thousands of languages to improve the quality of its output.
The technology powering modern AI translation systems is impressive, but the challenge it is taking on is immense. The same technology is set to identify illnesses before symptoms even develop; power driverless vehicles and “see” the contents of images.
Meanwhile, the output from the best AI translation tools available remains comparatively underwhelming. This does not make the technology any less impressive; it simply reinforces the reality of how much the human brain is doing when it listens or reads and understands language with all of the context and nuances we use on a daily basis.
Before we even think about translation, the prospect of algorithms matching humans’ ability to perform and understand a single language exactly is still science fiction.
To better understand the role AI translation will play in the language industry, it helps to look at how businesses are currently using the technology. Here are some useful examples:
Machine translation plus post-editing (MTPE):
Professional translators review and post-edit the content produced by AI translation to bring it up to the required quality.
For scenarios where there is no time for a complete translation process and AI is preferred to no solution at all.
International companies may need to access information in foreign languages and AI technology can provide instant translations (verify accuracy before drawing any conclusions or informing decisions).
For content that is not viewed publicly such as research, quotes, stats, news stories, etc. for internal use.
Instant translations can help teams devise strategic plans and respond to opportunities faster – again, translation accuracy requires verification before any action is taken.
Multilingual companies use AI technology to provide an automated (albeit, less than perfect and more mechanical) interaction with users.
The likes of Twitter and Airbnb use the technology to automatically translate user content, such as reviews, messages and posts.
International companies can provide instant responses to customer issues until support team members are ready to intervene.
Essentially, artificial intelligence translation steps in when time is short and you need an instant response – i.e. when speed is more important than accuracy.
One of the best examples would be a multilingual chatbot embedded on a website, providing instant responses to user questions. The bot’s performance comes with limitations and the translations may not always be 100% accurate, but it is essentially filling a role that humans cannot.
The same is true for professional translators who use AI solutions to provide instant drafts, which they can review, edit and improve to achieve the required quality. If an artificial intelligence translation tool can achieve even 20% accuracy throughout a project, that is still a significant time and cost saving that makes a real difference to customers and for companies using these solutions too.
To learn more about how artificial intelligence translation can help you save time and money on your translation projects regardless of their scale, get in touch with us – fill out the form on our contact page.