For decades, machine translation was the stuff of science fiction, enabling characters from around the world and across the universe to communicate without language barriers.
Now, science fiction has become reality…almost. Today, machine translation is widely available and can be an incredibly useful tool for people and organizations alike, if it’s used appropriately.
Used unwisely, however, it can damage your brand reputation, offend customers and clients, and even lead to injuries, illness, and legal liabilities.
With that in mind, here’s our in-depth guide on how to use machine translation correctly, in order to maximize the benefits and minimize the risks.
What is Machine Translation (MT)?
The most basic definition of MT is simply an automated translation from one language to another that’s performed using computer software. But within this basic definition, there are different types of machine translation algorithms that use different computational methods to produce translated content.
There is one important characteristic that all types of machine translation in use today have in common, however- the systems involved are not capable of understanding language in the same way as a human translator would.
Let’s take a closer look at the different types of MT and how they work.
This is the oldest type of machine translation. Rule-based MT is based on “rules” that guide the conversion of text from one language to another. The problem with rules-based MT is that each language has countless “rules” with just as many exceptions. So, this type of MT requires intensive human labor to program. Theoretically, rules-based MT can be constantly debugged, with targeted rules to correct errors, but in practice, the sheer complexity of language limits its accuracy and usefulness.
Pioneered by Google Translate, statistical machine translation was the technological advance that made free, public translation services available to the general public. Instead of trying to program every rule and exception manually, the programs compare texts that have already been translated from one language to another to determine the most likely translation for any given sentence.
Statistical MT was a revelation. Suddenly, free, fast automatic translation was at the world’s fingertips. However, it was far from error-free, as businesses that tried to cut out human translators soon discovered.
For example, see our roundup of the funniest Google Translate errors here.
And now we come to Neural Machine Translation (NMT), the newest method of MT. NMT uses neural networks modeled after our own brains to produce more accurate translations than ever.
Will this be the innovation that finally removes humans from the equation altogether? Technologists certainly seem to think so, and the press has consistently echoed their claims. But a quick look at the history of machine translation illustrates how often that expectation has been dashed.
Machine Translation Vs. Human Translation
There’s no denying the advantages MT offers to organizations of all stripes: price, speed, and scale. With so much content being created, there simply aren’t enough human translators to go around. In fact, according to CSA research, “just 0.00000008% of the data generated daily is likely to be translated.”
Human-powered translation also takes more time- hours and days instead of minutes or seconds. And machine “labor” is cheaper than human labor- human translators are skilled professionals and they deserve to be paid accordingly.
However, human translators also have key advantages over MT: comprehension, creativity, and cultural understanding. And machine translation has disadvantages of its own.
The biggest problem, of course, is quality. Common problems with MT output inaccuracy and poorly-written copy. If not corrected, these problems can impact the readers’ experience and their opinion of your organization. For websites, Google penalizes bad machine translation, and that means low-quality MT can also impact your SEO.
Can MT Replace Human Translators?
For decades now, science fiction writers, journalists, and futurists alike have been predicting the day that computers take over the jobs of human translators and interpreters. If we can’t have flying cars, the thinking goes, at least we can have Star Trek’s universal translator.
Claims that MT had finally reached “human parity” started with Google’s unveiling of its NMT translation in 2016. In 2020, researchers affiliated with Google even claimed that they had developed an NMT engine (called CUBITT) capable of “outperforming” human translators on specific sorts of news articles.
As Slator points out here, the provocative headline camouflages much more limited claims within the paper itself. There, researchers noted that the CUBITT engine produced translations that evaluators rated higher than human translations on measures of “adequacy”. With that said, human translation was still rated as “more fluent.”
“But Can’t I Just Use Google Translate?”
But even before advances in neural machine translation, translators had been hearing the question “Can’t I just use Google Translate?” repeated over and over again.
When comparing a human translation to MT, it’s important to consider the quality of both the MT platform in question and the translators involved. A bad translation is a bad translation, whether it’s produced by a human or by a computer. It’s certainly possible for a sophisticated MT system to produce equal or better results than a poorly trained human who isn’t quite fluent in one of the languages involved.
But when you’re comparing quality translators to MT, humans generally win the day. Skilled human translators have an understanding of the languages and cultures involved, and how those languages and cultures interact, that computers simply don’t have. For more on how humans perform when compared to NMT, see A Translation Showdown: Man Vs. Machine. Spoilers- the machines lost that round!
At the moment, machine translation works best when it’s combined with human post-editors to ensure accuracy and fluency in the finished product.
Machine Translation Platforms
Organizations considering MT have a variety of solutions available, from big names in tech like Amazon, Google, and Microsoft as well as from companies that focus specifically on MT, like DeepL.
The quality of the output you get will depend heavily on which platform you choose. So, which is the best?
That’s where it gets tricky- there’s not one MT platform that’s going to be the best for every organization and every type of content. Different platforms provide different levels of accuracy depending on a variety of factors, including:
- Language pairs. In fact, one study found that 8 different platforms would be required to get the best possible quality amongst the top language pairs!
- Type of content.
- Customization options.
To learn more about the top MT platforms and how they differ in terms of these and other considerations, see “5 Different Machine Translation Platforms, Reviewed.”
When and How to Use Machine Translation
When deciding on whether to rely solely on MT, your organization must consider the type of communication involved and the potential impact of a mistake. Would an uncaught error harm your brand reputation? Would it violate applicable laws and regulations? Could it put your customers in danger?
Unedited MT is best for translating large amounts of content when accuracy is not paramount. For example, it’s a good option for translating user-generated content, like reviews or social media posts. It’s often used for legal e-discovery, to sift through large amounts of content to identify important documents that may need more professional attention. Other common use cases include internal company documents and product descriptions for e-commerce. But here again, companies must carefully consider the potential impact of translation errors, as well as data security for internal documents.
And if accuracy IS essential? Then it’s not a good idea to trust MT alone. With that said, MT can still be part of your workflow. When combined with post-editing by a qualified linguist, MT can save time and money without compromising quality for a variety of different content types.
When MT Isn’t Worth It
However, there are also situations in which MT doesn’t make sense at all. For example, highly creative marketing and advertising content often require more than a simple translation to preserve its intended impact and meaning across cultures. “Transcreation,” the act of recreating a campaign to appeal to another language/culture, is often the best choice for creative content.
Legal and Ethical Considerations for Machine Translation
It’s also important to consider the legal and ethical ramifications of using MT, for both your business and your customers.
Errors generated by MT can be a compliance nightmare if not corrected. For example, consider product labeling, where accuracy can literally be a matter of life and death for customers with allergies. And even if nobody is harmed, inaccurately labeled goods may be recalled or impounded at the border, at significant expense to you.
If you’re a healthcare provider or involved in public health, Google Translate can seem like a godsend. But it can introduce dangerous inaccuracies. For example, the Virginia Department of Health website relied on Google Translate to provide information on the COVID-19 vaccine in Spanish. The resulting text told Spanish readers that the vaccine was “not necessary.”
The potential for mistakes isn’t a reason to forgo the advantages of machine translation entirely, but it’s one more reason to insist on a comprehensive quality control process.
Privacy and Security
Another potential issue is privacy and security. If you’re handling confidential data of any sort, it’s your job to ensure that data stays protected.
That means making sure that your MT platform is using proper data security protocols. If you’re going through an LSP, ask about their practices, procedures, and certifications. For example, if your company is bound by HIPAA, the law requires that business associates, including translation companies, be HIPAA-compliant, too.
If you’re using MT on your own, use a paid platform and make sure that platform keeps your data secure.
Public, free MT options may not safeguard your data and have been implicated in breaches. For example, in 2017, Statoil, Norway’s state-run oil company, was found to have been using Translate.com for documents containing sensitive information, including staff termination letters and employee performance reports. As a result, until the breach was discovered, all of this information was freely available online to anyone who knew where to look.
How to Use Machine Translation
To get the most out of MT, there are some “tips and tricks” you should know.
- Customize your MT platform to improve accuracy, if possible.
- Prepare your content for MT: Keep sentences short and simple. Use active voice instead of passive. And don’t forget to spell check and proofread.
- Robust quality assurance. Have important content post-edited by a qualified translator to catch errors and mistakes.
For more information, see How to Use Machine Translation: Our Top Tips.
The most important tip of all? Unless you have the team on hand to measure every aspect of a translation/localization project, you’re better off partnering with an experienced LSP to use MT effectively.
Why Use An LSP?
An LSP can:
- Identify where MT would best serve your interests.
- Select the best MT platform and customize it.
- Integrate translation memories for more consistent translations,
- Post-edit to ensure the resulting content is error-free and in line with your organization’s style guide.
- Ensure that all data is handled securely from start to finish and that the final product complies with all applicable regulations.
- Ensure that translated text displays properly and seamlessly handle any necessary design changes.
At K International, we can integrate MT solutions to save our clients time and money, while still ensuring accuracy and compliance. We combine translation memories from previously approved translations with an MT engine. Once the computers have worked their magic, a post-editor checks it against the source and makes any necessary changes. After that, it’s sent to another linguist for proofreading, and then to the project manager for another round of QA.
Should your needs go beyond machine translation, we also have a skilled team of multilingual copywriters, graphic designers, designers and specialists to provide comprehensive support for all types of localization projects.
For more information on how we can help, contact us today!