When the idea to use machine translation for Subway’s content surfaced, their Globalization Services Manager, Carrie Fischer, said: “There’s no way this is going to work for Subway’s content.”
A little later, she chose to give it a go. This is what she saw:
Savings based on four languages: $172,372 (or 44%) and at least 2 months of time.
In just one year, Subway saw an over 1,000% return on their investment.
Many of us have been here before. Between the media hype and exaggerated claims, sometimes it’s hard to distinguish when machine translation will actually be good enough for most use cases.
But in the past five years, we’ve seen neural machine translation engines like Google Translate (GT) really take off. Even if machine translation isn’t yet perfect, it can already help humans translate much faster.
Here’s everything you need to know about GT, and when it’s worth considering for your business.
How accurate is Google Translate?
We won’t bore you with how GT works. You want to know how it can work for you. Okay. So how accurately does GT translate content?
In a study carried out in part by the UCLA Medical Center, they found that “overall, GT accurately conveyed the meaning of 330/400 (82.5%) instructions examined but the accuracy varied by language from 55 to 94%.”
Here are the accuracy numbers for the following target languages using English source content:
| Spanish | 94% accurate |
| Korean | 82.5% accurate |
| Mandarin Chinese | 81.7% accurate |
| Farsi | 67.5% accurate |
| Armenian | 55% accurate |
As you can see, there’s a wide accuracy range for different target languages. Because GT supports so many languages, its accuracy depends on the specific language pairs you’re targeting.
For example, translating from English to Spanish has a very high accuracy rate (94%) because it’s such a common language pair. In contrast, translating from English to Armenian has a lower accuracy rate (55%) because it’s not nearly as popular.
Is DeepL better than Google Translate?
DeepL and Google Translate are the most popular generic machine translation (MT) engines on the market. DeepL is currently available in 29 languages, while GT is available in over 130.
Research from Intento shows that Google and DeepL rank highest (out of 18 generic engines) for more or less all language pairs, with Google taking the lead:
How do these providers actually perform when it comes to some of the most common language pairs? We found a study done by DeepL looking at the following language pairs:
- English → German
- German → English
- English → French
- French → English
- English → Spanish
- Spanish → English
While it’s clear that the accuracy of Google Translate is much better than both Amazon and Microsoft, DeepL seems to lead in the above language pairs. Of course we’re aware that a study not conducted by a third party can lead to bias.
We cross-referenced these results with the Intento report and found that DeepL ranked higher for German, French, and Spanish translations, while GT was better for Arabic, Korean, Brazilian Portuguese, and Mandarin Chinese.
Aside from language pairs, Intento also analyzed performance in different content domains such as software localization, ecommerce, retail, fintech, and medicine. Both DeepL and Google were the main contenders with very little competition from other generic MT engines.
Note: While DeepL and GT are clever enough to analyze the context in which a word is used and adjust the translation result accordingly, they sometimes fail to accurately translate isolated terms.
For example, check out this hilarious result of a French → English translation using DeepL and GT:
Input word | DeepL | Google Translate |
eau de toilette |
DeepL failed to accurately translate ‘eau de toilette’ as there’s no way it could know whether this word is needed to describe a perfume or for a plumber’s guide on fitting a toilet.
Why you should use Google Translate
GT is darn clever. And we must admit, it’s a lot faster than us humans. According to Konstantin Dranch, Founder at CustomMT, a translator can boost output from roughly 2,000 to 4,000–8,000 words per day while maintaining a good level of quality, if she (or he) switches to post-editing of machine-translated texts.
“A huge increase in productivity after a certain period of adaptation is the reason that machine translation post-editing (MTPE) is quickly overtaking regular human translation,” says Konstantin. At Lokalise, some of our Lokalise customers use GT combined with translation memory for pre-translation, testing, and post-editing.
The reasons for using GT are obvious – you save both time and money, but here’s a more specific/comprehensive list of the benefits:
- Reduce costs by 30–70%
- Faster turnaround for high-volume translations (a translator can double output while maintaining good quality if he or she switches to MTPE)
- Increase the number of supported languages
- Reach zero backlogs in the localization department
- Increase the share of raw MT projects vs. human translation or MTPE
- Help translators achieve equally high productivity and transparency on their MT usage
When you should use Google Translate
Machine translated content has a place and time. Your brand can’t do without human translators, so we’ve categorized content into three buckets that will give you a clear overview of when to use GT:
Low visibility + low importance → Machine translation
Not all content is created equal. You might not always need the highest quality; you need what will work for you according to the intended use.
Content like internal documentation, customer support articles, and FAQs don’t require perfect style and consistency. They need to be effective, meaning your customer can access the necessary information to solve their problem.
For most content of low visibility and importance, GT will likely be sufficient to get the job done. We’ll cover how you can incorporate automated QA into your workflow below.
Pro tip: If you use GT, you should own it. Here’s a tiny hack to avoid frustrating your users with GT:
All you need to do is add a little explanation or tooltip at the top of the page. If your readers can detect your copy has been machine translated they’re likely going to bounce off the page or complain.
High visibility + high importance → Human translation
For high visibility/high priority content like website copy, marketing or advertising copy, and sales collateral, you’ll want to be sure that your company’s distinct style and voice comes across just right.
Marketing localization is an exercise in technique and creativity for translators. Writers have to cram the intended message (the essence of the text) into a few words, and those words have to be just as impactful for the target audience as they are in the source text.
Most people wrongly assume that short texts need less time to be translated – it’s “only a couple of words.” But that couple of words can make or break a promotional campaign if not translated properly.
All other content → Machine translation + human translation / post-editing
For (most of) your other content, the key is to balance your budget, technical capabilities, and quality standards.
Do you have a low pass score for language quality to meet your company’s standards?
If so, you can use GT and in-house resources to get feedback from internal stakeholders.
Are you primarily translating simple text such as buttons or other product UI needs?
If yes, then GT is a cost-effective solution that can work well when combined with partial human QA.
Prioritize time in areas that build foundations, such as optimizing tooling, and areas that evaluate the results of that work, like receiving feedback.
How to use Google Translate with your translation management system
Cloud-based translation management systems (TMS) have made it easier than ever to integrate MT into your translation workflow while ensuring quality translations.
Here’s an overview of the steps involved in a GT + TMS workflow:
As you’ve likely noticed, people who work in the localization and translation industry use loads of jargon and abbreviations. Before we get into how you can build a translation workflow with GT in Lokalise, let us clarify the acronyms used in the workflow above:
Content management system (CMS): A system where content is created, stored, and published.
Translation management system (TMS): A system where multilingual content is created and stored along with the source language equivalent. Take a look at this post to get a holistic view of a TMS and how you can choose one that works for your team and company.
Translation memory (TM): A database of sentences, or segments of text, and their translations that can be automatically reused when translating similar or identical content. Everything that you (or any other team members) type in the editor, upload, or set via an API is saved automatically for future use.
LQA: A process used to intake bugs and feedback and ensure that all critical issues are fixed by launch. Take a look at this post to learn how you can build a linguistic quality pillar that ensures quality standards at scale.
In an ideal world, your content will live in a CMS with existing APIs or connectors to your TMS. While it’s fairly common for companies to have multiple systems for managing content, often the TMS is one of the only places that is truly “central” company-wide (where all of this content gets stored, in multiple languages, along with the source language equivalent).
Because localization is a team sport, the more people you require to manually handle different steps, the more it will cost. In light of this, here are some examples of what your TMS should help you automate:
1. Transfer of files between systems (e.g., two-way integrations that sync with code repositories, CMS, and design tools)
2. Conversion of files to different formats
3. Version control to quickly detect file updates
4. Automatic translations using machine translation
The above is by no means a comprehensive list of the automation capabilities your TMS should provide, but in general, that’s what you’ll need to automate as much as possible and minimize human work.
To start a machine translation project in Lokalise:
1. Choose Google Translate as your machine translation engine.
2. Upload the file you want to translate.
3. Start translating. You will see the MT suggestions, and the MT result will be automatically inserted into the text box where you can edit it.
Note: Everything that you enter from MT in the editor is automatically saved into the translation memory for future use.
4. You can post-edit the machine translation suggestion, rewrite it entirely, or simply accept it. The best chance you have of increasing quality is by fixing it at the source. Your TMS should help you do precisely that using built-in QA checks.
5. Once you’re done translating a key, Lokalise will automatically save this translation option to the translation memory. Later, when you’re translating the same phrase for the same language, Lokalise will give you a handful of inline suggestions from the TM.
Pro tip: If you’re using design tools like Figma or Sketch, you can use our integration and populate your designs with MT to check how the design changes based on the target language(s). This is a simple way to test the waters with design-stage localization (a powerful way to continuously release fully localized products like mobile apps, web apps, and games).
To Google Translate, or not to Google Translate?
GT is a great addition to your translation team. Every night, when your translators are asleep, GT takes up work, goes through your database of texts, and delivers them in your target languages. When your human translators arrive in the morning, they can focus on editing and working on key content that is highly visible and important.
As Ray Kurzweil, Director of Engineering at Google explained: “These tools are going to increase our ability to use, create, understand, manipulate and translate language… The idea is not to resist the tools, but to use them to do more.”
FAQ
1. Are there better machine translation engines than Google Translate?
While GT is one of the best generic machine translation engines, there are custom engines that are better for specific content domains and language pairs. That said, we always recommend that you combine MT with human review.
2. How do I build a review step to ensure human review of automated translations?
How you set up your review team will depend on your resources. Generally, the options you have available fall into two buckets:
- In-house: Internal company resources like native speakers on your team or in-country resources.
- Outsourced: Freelance reviewers, local subject matter + language experts, or your language partner
If you’re working with a language partner, the ideal scenario is to have them take care of translation review using a TMS. Lokalise makes it easier to manage translation reviews by providing the necessary functionality to automate a chunk of your review process. To expand QA checks and spellchecker languages, you can install lexiQA’s chrome extension for Lokalise.
