A quick overview of neural machine translation

Neural machine translation (NMT) has been getting a lot of attention in recent years given its effectiveness in language translation and localization. This translation method uses deep learning technology to not only translate text, but also improve the accuracy of its translations over time. It’s no wonder that it is one of the most accurate translation technologies in the market right now.

If you run a business, using neural machine translation to translate your website is a smart strategy for reaching new markets, optimizing your website for multilingual searches, increasing international sales – and ultimately, growing your business. While neural machine translation can sound complex, leveraging it for website translation can be surprisingly simple. What’s the secret? You’ll find out as we spill the beans in this article!

Read on as we explore how neural machine translation came into being and provide a beginner-friendly explanation of how it works. We’ll also shed light on how you can use it to translate your website without needing to first obtain a Ph.D. in machine translation!

What is neural machine translation?

To explain what neural machine translation is, we need to first have an understanding of machine translation (MT). Simply put, machine translation is the process of using computer software to translate text from one language to another. Feed your input sentence to the machine translation software, and it will automatically generate translated text in your target language for you. No human involvement needed.

Machine translation is used in all sorts of ways, such as:

• Localization of software applications by developers.
• Translation of website copy into multiple languages to cater to a global audience.
• Translation of subtitles for television and movie consumers.
• Translation of patent documents before filing them with patent offices worldwide.

Machine translation techniques have evolved over time, with neural machine translation being the latest and greatest version of the technology. Compared to traditional machine translation, neural machine translation uses artificial neural networks to translate text.

Don’t worry if this sounds confusing. Coming up next, we have a quick history lesson on the development of neural machine translation – plus an explainer of how this advanced machine translation technology works.

How was neural machine translation developed?

The earliest form of machine translation was rule-based. Such software would analyze the source text word for word, then refer to a set of rules developed by linguists to decide how to translate each word. Simple rule-based translation can be dated as far back as the Cold War era, when Americans used it to translate Russian.

But translating words one by one using such a crude system didn’t give rise to the most accurate translations. After all, some source texts may be best translated in phrases or whole sentences at a go. In this regard, statistical machine translation (SMT) models – being the next evolution in machine translation – did somewhat better.

Statistical machine translation software would first crawl through massive datasets of human-translated texts (also known as bilingual text corpora). Then, it applied prediction algorithms to identify words and phrases in the source text and determine how to best translate them.

Such technology improved over time, essentially growing into neural machine translation as we know and heavily rely on now. We’ll cover how neural machine translation works in more detail in the next section.

How does neural machine translation work?

Without getting too technical, neural machine translation uses deep learning techniques and artificial intelligence to:

• Analyze the source sentence, and
• Find patterns for generating a suitable translation in the destination language.

This technology is based on deep neural networks, which are an interconnected series of neurons or “nodes” modeled after the human brain. There are various types of deep neural networks, but one example would be recurrent neural networks, or RNNs, which may use an encoder-decoder architecture with an attention mechanism.

Before being deployed for translation, the neural MT software will be given training data in the form of different examples of translations for a certain text. Using such data, the software is then “trained” to produce the most accurate translation for a particular situation.

Why is neural machine translation so important, and what makes it better than other translation techniques?

Neural machine translation has many advantages over traditional machine translation. These include:

Higher translation accuracy

Previous traditional machine translation methods weren’t sophisticated enough to translate certain especially complex languages – in the sense that the resulting translations were so poor, they were practically unusable without first undergoing major manual revisions by humans.

But with their capability to “learn” over time, NMT systems are constantly improving the quality of their translations. This is unlike traditional machine translation systems, which have no function for “self-learning” and adapting their translation output over time. Accordingly, when sufficiently trained, neural machine translation software can produce much more accurate translations compared to their traditional counterparts.

For example, Google previously found that its Google Neural Machine Translation (GNMT) system was able to reduce translation errors by about 60% compared to its phrase-based production system.

Similarly, we recently conducted a study on the usability of machine translation for website translation needs. After reviewing the quality of website translations produced by various leading NMT technologies, we found these translations to be highly usable and require at most minor editing.

The neural machine translation technologies also particularly excelled in translating German, generating the most segments of “no-touch” translated text (that is to say, they did not require manual editing).

Less need for human input

After source text has undergone a first pass of machine translation, it usually goes through further human refinement to ensure its accuracy and suitability for the target audience.

The higher translation accuracy provided by neural machine translation means that the resulting translations need less manual adjustment (also known as “post-editing”) before being ready for use.

Quicker translation turnaround times

When businesses can obtain more accurate machine translations that require less post-editing, they’re able to start making use of the translations sooner. But apart from that, it’s possible to train neural machine translation models within a short period of time. This in turn allows for speedier translation processes.

For example, Facebook uses neural machine translation to translate text in posts and comments (and as you probably know, there is a lot of such content on its platform). While the business previously needed almost 24 hours to train its neural machine translation models, it was able to cut this timeframe to just 32 minutes!

Can you apply neural machine translation to your business and how?

Using neural machine translation to translate your website may sound intimidating, requiring a costly investment into new tech and research and development. This is not true! Nowadays, there are many pre-built NMT tools available in the market to help you translate your website content into various languages. These tools are also quite affordably priced. In fact, they tend to be cheaper than engaging a professional human translator to translate your website in its entirety.

Our Weglot website translation solution is one such tool. We use a proprietary mix of NMT translations from leading machine translation providers DeepL, Microsoft Translator, and Google Translate to instantly generate translations that are higher in quality compared to using these machine translation technologies on their own. We support the translation of over 110 languages, from common ones like English, German, and Italian to less common languages such as Tatar and Malagasy.

Weglot also does more than just text translation. Among other features, it can:

• Translate images and videos: Weglot provides a built-in media translation feature for displaying different images and videos to audiences from different countries. It works by letting you specify alternative media URLs for different language versions of your website content. You’ll find this feature especially handy where your media contain wording or need to be swapped out to fit the local context.
• Facilitate team collaboration: Roping in staff members or external translation agencies to review your translations is straightforward in Weglot. All your translations are stored in the centralized Weglot Dashboard, and you can add collaborators by their email addresses to grant them access to your translation project. You can also adjust their access permissions anytime as the situation requires. You’re in complete control.
• Optimize your website for multilingual searches: Multilingual search engine optimization (SEO) refers to the taking of steps to enhance your website’s visibility in multilingual search engine results. It’s an essential process if you’re building your website with a global audience in mind. With features such as automatic hreflang tag implementation and metadata translation, Weglot will help set your website up for success in the multilingual search arena.

Weglot offers no-code integrations with leading website platforms such as WordPress, Webflow, and Shopify. It’s easy to set up, and can help your business:

• Save money: Let Weglot produce a first pass of machine translation of your website content. If needed, you can then have a professional translator look over and adjust certain key translations before you use them on your website. Employing this language translation strategy can save you a lot of money compared to having a professional translator translate all your website content from scratch.
• Save time: Manually translating text word for word not only leaves room for human error, but is also a very slow process. Weglot is capable of translating large volumes of text in a short time, helping you make your website multilingual in as quickly as a matter of minutes.
• Increase the quality of your translations: As mentioned, Weglot uses a proprietary mix of neural machine learning translations to enhance the accuracy of translations. This certainly beats trying to translate text on your own – especially if you aren’t a skilled linguist in the relevant language pair. We’re also constantly refining our translation technology to help you get nothing but the best results!

More than 60,000 websites use Weglot for their translation needs, with outstanding results. French eyewear brand Jimmy Fairly is one of them: the business doubled its percentage of international sales just one week after using Weglot to translate its French website into English. Encouraged by these results, Jimmy Fairly subsequently translated its website to German and Italian as well and reaped an overall 70% increase in web sessions after eight months.

Tap on the power of neural machine translation for your business with Weglot

Delivering superior translation performance, neural machine translation is a significant improvement over other traditional machine translation techniques. With that in mind, it’s unsurprising that NMT is the machine translation technology of choice for so much of the business world.

As our machine translation study reveals, the volume of machine-translated-web content has increased sixfold over the last two years. Machine translation is also used for large projects, with more than 10% of translated websites containing over 50,000 words. Finally, only about 30% of machine-translated content is edited, suggesting that a significant portion of machine translations is so accurate that it requires no further refinement.

Weglot makes it easy to harness the power of NMT to translate your website content. It’s simple to set up and automatically chooses the most suitable machine translation engine to produce the best translation for any given language pair. The result is speedy, high-quality website translations that you can implement without fuss. Just take it from Corrine Ellsworth-Beaumont, the CEO of Know Your Lemons Foundation:

“With Weglot we were able to translate a number of pages into 10 different languages within the first week of the website launch, dramatically increasing our international visibility.”

If you’re keen to experience the power of Weglot for yourself, you can sign up for a free trial here.