Machine Translation is often seen as an easy, low-cost fix for the complicated problem of translating digital and offline content at scale.

There’s good reason for this; after all, innovative approaches—such as Neural Machine Translation—are getting better every day and can iteratively improve their translation quality over time. And since they technically require no human involvement to generate translations, machine translation is substantially more affordable than human translation.

In fact, there may be a day within the decade or two when machine translations appear so seamlessly and authentically human that professional linguists won’t be needed at all.

We’re a long way from that possible future … and in the meantime, the revolutionary technologies powering machine translation aren’t quite “there” yet. It’s not the answer for all translation use cases.

This section takes a closer look at the evolution of machine translation and addresses several topics to consider as you examine options to translate your website, such as:

The Innovations of Neural Machine Translation (NMT): NMT uses a complex computer “neural network” that processes information in a nonlinear way, like the human brain does. Thanks to this breakthrough, NMT platforms deliver far more accurate, nuanced translations than other NMT technologies.

The Benefits of NMT: We’ll take a close look at the many upsides to using NMT for website translation including its low cost, rapid output and low-risk applications in new markets. NMT can “learn” too, enabling companies to further improve translation quality.

The Challenges of NMT: NMT has some drawbacks, and we’ll explore them. For instance, they tend to fall short when presented with specialized content, such as technical, scientific or legal material.

NMT and International SEO: We’ll also offer a word of caution about depending on NMT to effectively optimize your translated website for organic multilingual search. (Spoiler alert: It’s not recommended.)

Best Applications of NMT: Saving the best for last, we’ll present a clever way to maximize the effectiveness of NMT for website localization while ensuring pitch-perfect translations where you—and your multilingual website visitors—need it the most.

The History and Common Uses for Machine Translation

Machine translation—sometimes called automated translation—is the application of software to translates text from one language to another without human involvement. The linguistic quality of machine translations has dramatically improved in recent years … and especially since 2016, when Google debuted its groundbreaking Neural Machine Translation system.

At their core, machine translation systems don’t comprehend languages the way humans do. Instead, their translations are generated by probabilities—best guesses—to determine which translated words and phrases should be used .

Since machine translation requires no human effort to generate translated text, it offers several advantages for organizations that wish to use it, such as:

  • It’s more affordable than human translation
  • Content—such as hundreds of webpages or offline documents—can be translated rapidly
  • Using newer machine translation solutions, translation quality can improve over time by “teaching” the software with additional content

Machine translation also has its drawbacks, mostly because it doesn’t fully understand how human beings actually speak to each other.

  • Machine translation often generates translations that reflect simplistic word choice and fluency, particularly when compared to human translations
  • Software is often unable to account for regional or dialect variations
  • Unknown words, such as idioms, usually stymie machine translation systems

There are ways to mitigate these shortcomings, though they often require additional time, effort and resources such as:

  • Humans can edit machine-translated content after it’s been generated
  • Some machine translation approaches can be iteratively “trained” with bilingual content to improve translations
  • Machine-translated content can be applied to use cases where nuanced linguistic fluency isn’t required, such as product specs and descriptions

Let’s take a closer look at the different kinds of machine translation available to organizations today.

Types of Machine Translation

There are three approaches to machine translation:

  • Rule-based Machine Translation
  • Statistical Machine Translation
  • Neural Machine Translation

Rule-Based Machine Translation

Rule-Based Machine Translation applies a specific set of programmatic rules, manually created by developers and linguists, that map a dictionary of translated phrases and words to comparable phrases and words in a source language.

These rules can be adjusted over time to improve translation quality.

Due to the high involvement of programming and manual word selection, Rule-Based Machine Translation generates more predictable translation results than other methods, such as Statistical Machine Translation. While this predictability in word choice may make translations easier to revise by editors, the output is often:

  • Grammatically unsophisticated
  • Linguistically stilted

Further, building dictionaries for these systems is time-consuming and costly. Programmatic rules are also hard to make for ambiguous phrases and idioms.

Statistical Machine Translation

Statistical Machine Translation is widely considered a step up from Rule-Based Translation. It’s inspired by the theory that language has an inherent logic that can be mathematically solved.

The concept, at its simplest, is that a computational system identifies “conclusions” in the target translated language based on what already exists in the source language.

The process begins with a large data set of approved previous translations. This translated content is compared with content in the source language; translations are selected based on their statistical probability of being correct.

This generates more accurate translations that are far more fluid (and less predictable) than Rule-Based Translation. But this model also has shortcomings:

  • Longer sentences can be broken up into multiple sentences, while shorter sentences might be merged
  • The grammar of some languages defies conventional “word pairing” in a database, creating translations that are difficult to understand
  • Idioms, language-specific word order and “out of vocabulary” words (not stored in the database) also confuse Statistical Machine Translation systems

NMT: Out with the Old, In with the New

Both Rule-Based and Statistical Machine Translation are older technologies that recognize text at the word, phrase or syntax level. This approach examines a sentence’s content on a granular level but doesn’t account for the full contents—or context—of the entire sentence. Translations suffer as a result.

In contrast, the latest innovation in machine translation, called Neural Machine Translation (NMT), processes information similarly to the human brain: nonlinearly, using different layers of a neural network. This enables Neural Machine Translation platforms to read and translate content at the more holistic sentence level, resulting in far more accurate and nuanced translations.

Let’s dive a bit deeper into Neural Machine Translation.

Neural Machine Translation

NMT is the most advanced form of automated translation software available today. With recent advances in self-learning AI, deep learning and big data, NMT systems essentially “learn” new languages and apply linguistic knowledge to repeatedly produce translated content that’s more accurate than other machine translation approaches.

By now, every major provider of machine translation uses NMT. It’s fast becoming a viable, affordable way to translate online and offline content for multilingual markets.

Benefits of Neural Machine Translation for Websites

Thanks to recent technological breakthroughs, companies can benefit from NMT in several ways, including:


NMT translates content into widely spoken languages with far fewer errors than traditional machine translation.

High Volume, Low Cost

NMT can localize large volumes of content such as product pages and descriptions, offline documents or user-generated content almost instantly, for a fraction of the cost of human translation.

Low-Risk International Expansion

With NMT, companies can minimize their translation spend while testing the waters in new online markets. Successful localized websites can be enhanced with human translation as needed.

Flexible Application

Businesses with small budgets can affordably create localized customer experiences, with the flexibility to apply human translation to strategically critical content within spending limits.

Comprehensive CX Localization

Depending on your translation vendor, your company may also be able to apply NMT to other features, such as:

  • Localization of images, multimedia, web applications and third-party content
  • Localizing social media content, and real-time sentiment analysis of multilingual customers’ social posts
  • Instant translation for chat or email applications


Organizations can improve the translation quality of their NMT content by by “training” the database’s language patterns with additional content. This is accomplished by editing NMT output with a human translator, and re-teaching it to the system.

NMT systems are most effective when they have been trained with this bilingual content, and learn the linguistic rules needed to understand and interpret language patterns.

This is the best way to get the most accurate results that reflect a brand’s voice and messaging.

Challenges of Neural Machine Translation for Websites

Despite its ability to deliver far superior translations compared to other machine translation approaches, NMT isn’t without its flaws. Here are a few:

  • Lengthy “teaching” processes; training a NMT system requires a lot of energy and time
  • Inconsistent word usage, which can confuse readers or undermine a brand’s credibility
  • Inability to translate words outside the NMT database’s vocabulary
  • Unsophisticated handling of nuanced creative copy, such as wordplay

When it comes to digital content, translation takes on a secondary burden: International SEO. And while machine translation can be handy for translating a quick phrase for a piece of content, it’s just not ideal for handling full-scale website translation.

Here’s why:

Google’s own quality guidelines penalize the search engine rankings of websites that use automated content. This includes text translated by an automated tool without human review or curation before publishing.

So if you’re only employing solutions like Google Translate to present content in multiple languages on your site, you run the risk of Google itself penalizing the site for doing so. Lower search engine rankings can be crippling to companies that are looking to establish brand credibility and presence in new and unfamiliar markets.

Google also has pretty stringent quality standards for website architecture. That includes localized or translated websites.

This means that it should be clear to your visitors—and all search engines—how your translated content is built and organized so it’s easy to find, easy to navigate and easy to understand.

Here’s the problem:

  • Machine translation solutions like Google Translate can make that hard to do
  • They don’t actually create a “home” for translated content on your site
  • By simply skinning translated words over your existing content, machine translators avoid creating clear site architectures and sitemaps for content in multiple languages

Ironically, Google’s own machine translation solution falls short in one critical area: search engine optimization.

The Ideal Application of Neural Machine Translation for Websites

For businesses that want to leverage the cost-effective, time-saving benefits of NMT without risking their brand’s reputation with inconsistent word choice, the best translation vendors offer a hybrid approach, combining NMT with expert human translation. This delivers unmatched efficiency, accuracy and flexibility.

Companies can apply an NMT + human approach to use cases such as:

  • Using NMT for high-volume, low-traffic areas and human translation for brand sensitive or high-traffic areas such as homepages and landing pages
  • Translating all content with NMT, and applying post-editing with a human linguist on select sections or pages
  • Launching a digital localization project with NMT, and upgrading to human translation later, as business needs and budget allow

Translation technology experts predict that NMT is the future of digital localization. However, in its current state, more innovation is required before NMT overtakes human translation as the best and most accurate way to localize content.

The best way global businesses can use NMT is to leverage a hybrid approach that combines its cost-effective, time-saving benefits with the unparalleled linguistic, cultural and industry fluency of professional translators.

Read more about website translation in our ultimate guide to website translation.


Let’s Talk.

Ask us how we can help you grow globally, easily.

Schedule a conversation