How Accurate is Google Translate?
Is Google Translate accurate enough to count on for your business?
Easy. Instant. Free. These are some of the reasons Google Translate is one of the most popular translation tools. It definitely comes to the rescue when you’re trying to navigate a foreign country.
But, it can be unreliable for translating complex sentences for one main reason: it isn’t advanced enough to factor in all kinds of nuance and context. Still, it begs the question: how accurate is Google Translate? And can you depend on it for your website translation needs?
How Does Google Translate Work?
Ever wondered how Google Translate built its translation database? It’s heavily based on the Europarl Corpus, which is a collection of documents from European Parliament procedures that humans translated. Apart from that, it relies on various digital resources and common translations for languages.
When it first launched in 2006, it used statistical machine translation to provide instant translated text. But, it soon became clear that this wasn’t going to work in the long term. Since it translated individual words, it worked most efficiently for brief phrases. But it began to offer odd translations for longer, complex sentences.
Google knew they needed to swap machine translation (MT) technologies to improve its accuracy. 10 years later, in 2016, the company developed its own framework, Google neural machine translation technology (GNMT). The move made leaps and bounds in its algorithm and changed its approach to translation. Instead of translating each word, it looked at the meaning of the entire sentence.
The result? Translations that were much more faithful, even factoring in slang and colloquialisms. The change was massive: it cut translation errors by more than 55%-85% across many major language pairs.
With this new learning system, Google Translate stopped using English as a go-between for translating any language. It instead directly translated between two languages. That means it translated from French to Japanese instead of French to English, then to Japanese. By leaving out the middleman, it worked faster, more efficiently, and most importantly, more precisely.
Is Google Translate Accurate?
While Google Translate is available in more than 130 languages—making it a translation tool with the broadest range of support—it also varies in terms of accuracy rate. For instance, since Spanish is one of its most popular languages, its translation accuracy is typically over 90%.
In fact, a 2014 study found Google Translate to have only 57.7% accuracy when used to translate complex medical phrases. A 2021 study conducted by the UCLA Medical Center found that Google Translate preserved the overall meaning for 82.5% of the translations. But the accuracy between languages spanned 55% to 94%.
Sometimes, Google Translate’s precision is shockingly good. Results from our own study on the state of machine translation for website translation showed that 10 out of 14 translation editors were positively surprised by the quality of the translation they were shown. That means machine translation worked better than they expected.
One thing to keep in mind about the accuracy of Google Translate is that it works best when translating literary text into English. It’s a different story when it comes to informal phrases: it showed a 72% accuracy when converting English casual texts into other languages. That’s why you can end up with weird translations that don’t make sense when trying to render everyday expressions into other languages.
Is Google Translate Reliable for Website Translation?
Of course, efficiency is the name of the game here, and not everyone has the resources to leave thousands of words in the hands of a human translator, nor the time. Which is where machine translation comes in.
Machine translation has come a long way since its beginnings in the 1950s. With the rise of deep learning and neural machine translation (NMT), it’s become more reliable than ever. Still, because it deals with something as fluid as language, it isn’t always perfect.
English translation remains Google Translate’s biggest strength. According to a 2013 study that evaluated Google Translate’s accuracy in data extraction from non-English languages, extracting translated articles typically took longer than with English-language articles.
In our study, one of the translation editors observed that if the MT did not recognize the context for a particular term, it supplied a general translation instead. It produced incorrect translations due to the missing context. However, when enough context was provided, the translation proved to be accurate. But, can’t the same be said for human translation and dealing with text out of context?
The reason for all this is quite simple: Google Translate’s accuracy depends on how much data is available for the target language. Since 50.5% of websites are in English, Google Translate has a lot of input to work with. That’s why it has the highest accuracy for English language pairs. On the other hand, since only 2% of web pages are in Portuguese, Google Translate will have difficulty offering a highly precise Portuguese translation.
Even though a small percentage of websites may be available in a certain language, that doesn’t mean there isn’t a demand for it. 73% of consumers prefer product reviews in their native language when browsing online. If you’re hoping to expand your business across borders, it’s important to cater to native speakers of your target area. Website localization will help you nail your messaging and make sure nothing gets lost in translation.
Essentially, Google Translate is only as good as its users, or at least those who improve its translation quality. However, Google recently discontinued the 'Contribute' feature, where native speakers could provide relevant translation suggestions to improve the platform's accuracy.
While it’s a convenient base for translating your website, it works best when used with other translation tools and, of course, a human eye to look things over.
Are There More Accurate Translation Tools than Google Translate?
Other popular translation tools include DeepL, Amazon Translate, and Microsoft Translator, which all use machine learning to inform and improve their services. DeepL uses Linguee’s extensive database of manually translated phrases, sentences, expressions, and excerpts since the same businesses developed them. Amazon’s service relies on neural networks engineered to translate between a source language and a target language. Like Amazon, Microsoft Translator relies on NMT to power its translations.
Our study found that DeepL—which uses the same approach to machine translation as Google Translate—had the lowest number of unacceptable translations for Italian (it-IT). But it also had the lowest number of no-touch translations—which meant that it didn’t need a human to edit it—for the same language. DeepL also has the least number of supported languages at 28, but performed best at Spanish (es-ES) translations.
Meanwhile, Amazon Translate—which ranked the least-performing MT engine—outperformed its competitors in providing the highest number of no-touch translations in French (fr-FR). It also had the lowest number of unacceptable translations in Simplified Chinese (zh-CN). Its support is smack in the middle at 75 languages.
Microsoft Translator, which supports the second most number of languages at 111, performed fairly consistently. It did well with no-touch German translations, but it also had the lowest number of no-touch segments in Portuguese.
All tools performed well with European languages and, from a human editing perspective, provided high quality Arabic translations. In short, no translation software eclipses another—they work best when all used together.
Should You Use Google Translate?
Nonetheless, Google Translate is still a great tool that can save you a lot of time and money for your website translation needs. Accuracy highly depends on your intent, style, and the person receiving your message. So if all you need is to translate a few brief, simple phrases, it’ll serve you just fine.
Google Translate can be helpful in these instances:
- Editing low-visibility content, like website footers
- Instructional, repetitive content, such as guiding end-users toward vital information to solve a simple issue
- Product reviews, since there is a low expectation of high-quality, comprehensive feedback
On the other hand, Google Translate may not be appropriate in these cases:
- Translating everyday expressions that require more context or may not have a precise equivalent in your target language
- Using it as a dictionary to translate individual words, as it may offer a different meaning than the one the author intended
- Translating source text that may not have followed conventional grammatical rules in its own language or has syntax dissimilar to the target language—Like Spanish to Korean
- Translating sarcastic or ironic expressions
In these cases, since the quality of translation can make or break the message’s tone and delivery, having professional translation within your website translation workflow would certainly be beneficial.
The Best of Both worlds: Machine Translation and Human Editing
Google Translate is capable of many wonderful functions, but it shines best when working with other translation tools and human editors.
According to our study, 99% of translation work produced in the world is not done by professional human translators. And an average of only 30% of machine-translated content is edited. That’s actually a pretty big win for those looking at using machine translation. Though a human editor is still necessary to catch inaccuracies—artificial intelligence isn’t perfect—there’s a lot of value in using translation software. That means website translations performed by MTs are most definitely usable and require only some tweaking here and there.
It sounds like a lot of work, especially for small businesses. The good news? A website translation solution like Weglot makes this easier.
More than 110,000 global brands use Weglot to translate their websites using the best of machine translation and automatic translation. Though the two sound similar, automatic translation includes the entire workflow of proofreading, editing, optimizing the content for SEO, then re-uploading the content back to the website. That way, you’re guaranteed accurate translations.
Weglot works by choosing the most fitting MT engine for a specific language pair to provide the most precise outcome. Thanks to neural machine translation, Weglot detects and immediately translates all the content on your website. No need to painstakingly go through each page and translate them all by yourself. Best of all, it also displays those translations under separate language versions of your website.
It’s intuitive to use and allows you to translate and publish your content right away. So it’s even easier to localize and optimize your content in more than 100 different languages—even RTL ones like Hebrew and Arabic.
Curious about how it works? Try Weglot’s 10-day free trial (no commitment) to see just how easy setting up a multilingual website can be.