Machine translation (MT)—automated translation of text from one language to another via computer software—has been around longer than many realize. As with much of the technology we use today, MT has advanced significantly over the decades since its inception.
Neural machine translation (NMT), in particular, has become so good that it’s no longer a matter of whether you should use MT. The questions now revolve more around when and how. Before we get into the specifics, let’s take a brief look at the history of MT.
Machine translation: Past vs. present
Here’s a long story made short: MT started its infancy conceptually in 1949. Researchers struggled for years to make it a usable solution able to compete with human translators. The challenges of doing so were so great that, in 1966, there were talks about stopping funding for machine translation research altogether. The research continued, however.
In the '90s and early 2000s, MT (e.g., Babelfish and Google Translate) made it to the mainstream as the Internet started to take off. Then, in 2013, modern machine translation began gaining ground, starting with statistical MT and leading to neural MT.
- Statistical machine translation engines learn through statistical analysis of bilingual text. They essentially complete translations by predicting the probability of the next word in a sentence. While this method of MT was an advancement at the time, it had its shortcomings in terms of accuracy.
- Neural machine translation engines—the most popular way to do MT right now- leverage neural networks that mimic how the human brain works. Unlike statistics-based engines, they can consider the context of a sentence to translate it more accurately. They even learn over time, enabling progressively higher-quality translations.
Today, we see neural machine translation taking off, pushing boundaries some previously thought MT never could. According to Jack Welde, Smartling’s Founder and CEO: "Neural machine translation is gradually eliminating the demarcation between human and machine translation. It is creating more opportunities for a productive closed loop between machine and human, including machine-enabled tools that make the human more productive, and human inputs that make the machine more accurate going forward.”
This raises the question: When should you opt for MT over human translation?
When to use machine translation: Popular use cases
Complete human parity hasn’t been achieved yet—machine translations don’t perfectly match human-level quality. But modern MT engines are as close as they’ve ever been, meaning they can be useful on more language translation projects than was previously the case.
Before, when choosing between human and machine translation, the former would be recommended for any content that was high-priority, would have high visibility, or that had a lot of technical or industry-specific terminology. It’s no longer so cut-and-dry. Among other things, you can now use MT for the following:
- Marketing materials
- SEO content
- Product information
- User manuals
- Support tickets
- Knowledge bases and training documents
- User-generated content
- Commercial contracts
- Patents
- Terms and conditions
Doing so allows you to take advantage of its greatest advantage: speed. However, you can—and, in some cases, should—still have professional translators involved in your translation process. Post-editing is recommended for more critical content types. Think of things like contracts, which require total accuracy, or marketing materials since they require creativity to fully capture a brand’s personality and style.
Two machine translation tools that deliver guaranteed quality
If you’re considering incorporating MT into your workflows, what tools should you use and why? Smartling offers a number of machine translation options—powered by LanguageAI™ technology—to enterprise customers. These solutions produce translations that are up to 350% higher quality instantly and at a fraction of the cost of alternatives.
Smartling Translate
You can instantly and securely translate text and files in this AI-powered translation portal—no setup or training necessary. Smartling Translate leverages artificial intelligence—specifically, the GPT large language model—to improve the fluency and overall quality of translations. For example, in our GPT and Large Language Models Reality Series episode, we demonstrated how it can be used to pre- and post-edit to correct mechanics errors, adjust formality, and more.
Image Description: Smartling Translate translation portal
Neural Machine Translation Hub
Next is our Neural Machine Translation Hub, which is a private and secure AI-powered workspace designed for translation at scale. It can translate billions of words nearly instantly via the best machine translation engines:
- Amazon Translate
- CrossLang
- DeepL
- Google Translate
- Microsoft Translator
- Unbabel
- Watson Language Translator
- GPT
- Prompsit
- CustomMT
- PROMT
Even better, Smartling's AutoSelect can automatically route your content to the engine that will deliver the best translation. This results in up to 350% higher quality than a single-engine approach.
Image Description: Machine translation engines available in Smartling NMT Hub
Alternatively, you can train custom engines on your translation memory and glossaries for enhanced consistency across brand and industry terminology.
What to look for in machine translation software
As you decide how to incorporate MT into your processes, here are a few things you should consider.
1. Translation accuracyMachine translation quality estimation, which can be done manually or via automated assessments, can help you determine how well a certain engine handles your projects. Are its translations clear and accurate? Do they make appropriate use of industry terminology? Is the text free of biases related to gender, profession, or other characteristics? There are many considerations that can influence your quality estimates. But, ideally, you should use machine translation systems with a minimum of 90% human accuracy ratings.
2. Adaptive machine translation capabilities Besides rapid deployment, part of the beauty of modern NMT engines is customizability. While there are many out-of-the-box machine translation models available, translation quality will be best if the engine you use is trained on your data. An adaptive NMT engine will constantly grow its knowledge of your content, voice, tone, and overall style to deliver localized content that represents your brand accurately and resonates with your target audience.
Lucidchart is one of several Smartling customers who have enjoyed the benefits of adaptive NMT.
The company, which needed to speed up its translation process to match its fast release cycles, was able to reduce translation time from two days to just 10 minutes thanks to NMT. As a result, it’s now able to provide a consistent and cohesive localized product experience to its international users, without any negative impact on the product and release cycle.
3. Supported languages (and quality) Think ahead. You may only need to translate into a common target language like Spanish right now. However, if your company intends to expand into many new markets across the globe, chances are you’ll need less common or different languages (e.g., Asian or right-to-left languages). If you already have an MT solution that supports language pairs you may need in the future, that’s one less thing to worry about when the time comes.
A word of caution, though: Translation quality sometimes varies between languages and language pairs. So, it’s worth testing quality, not just availability. An even simpler alternative is using Smartling Autoselect, which takes this work off your hands and automatically routes your content to the MT engine that will deliver the best translation.
4. Integrations and automation Any good enterprise machine translation solution will not just translate your content but also streamline your translation workflows. Smartling Neural Machine Translation Hub, for example, has various capabilities. For one, it connects to our translation proxy. Smartling's translation proxy automatically ingests content from your website into Smartling for translation, then delivers the translated text to your site’s users instantly in their preferred language.
NMT Hub also integrates with top content management systems like WordPress, as well as various CRMs and low-code APIs.
5. Pricing What is your estimated monthly translation volume? The answer will determine which pricing structure will work best for you. For example, per-character pricing can be costly if you need to translate many long documents—per word or per document may be more cost-effective.
To maximize your translation budget, keep an eye out for tiered pricing with reduced rates for higher volumes. Also, ask about discounts for annual commitments or prepaid packages.
Just don’t forget that the actual translation isn’t the only thing that will incur a cost. You may have to pay to train a custom engine, integrate the engine with other tools in your workflow, or have human review and post-editing done. Tally up those costs, too.
Leveraging machine translation to deliver impactful experiences
Localized user and buying experiences are more important than ever. According to CSA Research, consumers strongly prefer to learn about brands’ offerings in their own language on localized sites. In fact, 40% or more of your total addressable market could be lost if you don’t acknowledge this preference.
While traditional translation certainly has its place, machine translation paves the way to deliver high-quality, localized content and experiences in a fraction of the time and at a fraction of the cost. You’d be wise to take advantage of it and, specifically, of neural machine translation tools like Smartling’s. Doing so will ensure you get nothing but the best quality. For more information on how you can leverage Smartling Translate, the NMT Hub, and our automation solutions, book a meeting with us today.
