Machine Translation Explained: Types, Use Cases, and Best Practices
The use of computers to translate text from one language to another has long been a dream of computer science. Nevertheless, it’s only in the past decade that machine translation (MT) has become a viable productivity tool in more widespread use. Advances in natural language processing, artificial intelligence (AI), and computing power all contribute to this increasingly useful technology.
To help you better understand its ins and outs, this guide will define machine translation and explain its types and benefits with many examples and tips. You will also learn about how to integrate MT into existing translation and localization workflows. Lastly, it outlines what makes the best machine translation software to support your workflows and help drive growth for your global business.
What is machine translation?
Machine translation is the process of automatically translating text from one natural language to another using a computer application. This means you add text to machine translation software in the source language and let the tool automatically transfer the text to the selected target language.
Unlock the power of machine translation
Discover advanced machine translation management features within our enterprise-ready TMS and create new business opportunities worldwide more quickly and efficiently.
The beginnings of machine translation
Translation was one of the first applications of computing power, starting in the 1950s with the famous Georgtown-IBM experiment. Still, the complexity of the task was far higher than early computer scientists’ estimates—requiring enormous data processing power and storage far beyond the capabilities of early machines.
It was only in the early 2000s that the software, data, and required hardware became capable of doing basic machine language translation. Early developers used statistical databases of languages to “teach” computers to translate text. Training these machines involved a lot of manual labor, and each added language required starting over with the development for that language.
Machine translation today
In 2016, Google implemented a key innovation in MT technology by shifting to a neural learning model, which was based on research from 2014. This approach involved training the MT engines using AI and proved to be far more efficient and faster than Google’s main statistical MT engine. It also exhibited remarkable improvements in translation quality as it was used.
Neural machine translation proved so effective that Google changed course and adopted it as its primary development model. Other major providers including Microsoft and Amazon soon followed suit, and the ever-increasing quality boosted the value of MT as an addition to translation technology.
Many translation and localization technology solutions now have integrated capabilities for machine language translation to help businesses meet the ever-growing need to overcome language barriers in the global marketplace. More on that later in this guide.
Download for free
Your up-to-the-minute guide to machine translation
Learn about new technologies to improve machine translation output quality, the latest on MT post-editing pricing models, and how to best shop for machine translation.
How does machine translation work?
Over time, machine translation development has yielded several types of machine translation systems, each with its own strengths and weaknesses. The 3 most common types of machine translation include rule-based, statistical, and neural machine translation.
Rule-based machine translation (RBMT)
The earliest form of machine translation, rule-based MT, relied on a large, predefined set of linguistic rules that helped the software transfer the meaning of text between languages. Overall, it had low quality translation, and it required adding languages manually as well as a significant amount of human post-editing.
Rule-based MT is rarely used today.
Statistical machine translation (SMT)
Statistical MT builds a statistical model of the relationships between words, phrases, and sentences in a given text. It applies the model to a second language to convert those elements to the new language. Thereby, it improves on rule-based MT but shares many of the same issues.
Statistical MT is mostly replaced by neural MT and is sometimes used for legacy machine translation systems.
Neural machine translation (NMT)
Neural MT uses AI to “learn” languages and constantly improve its knowledge, much like the neural networks in the human brain. As opposed to running a set of predefined rules, an MT engine’s neural network is responsible for encoding and decoding the source text.
Neural MT is more accurate, allows for adding more languages, and works much more quickly once trained—making it today’s standard in MT technology development.
Timeline of machine translation development history
Automated vs machine translation: The difference explained
Automated translation and machine translation are often confused to be the same, but they aren’t interchangeable terms as they serve entirely different functions.
Automated translation refers to any triggers built into a traditional computer-assisted translation tool (CAT tool) or cloud translation management system (TMS) to execute manual or repetitive tasks related to translation. It aims to make the overall translation process more efficient.
For example, automated translation may be used to trigger the machine translation of text as one of the many tasks in a translation workflow.
Machine translation is about converting text from one natural language to another using software. In other words, there’s no human input involved as in traditional translation. That’s why machine translation is also known as automatic translation.
Major machine translation providers
Leading developers of machine translation technology, like Google, Microsoft, or Amazon, currently use a type of neural MT as their preferred methodology—since it allows for both more nuanced translation and the constant adding of language pairs. This growth capability is made possible by the fact that machine translation engines can learn and improve as they are used more.
Machine translation engines work based on training data. Depending on your needs, the data can be generic or custom:
- Generic data is simply the total of all the data learned from all the translations performed over time by the machine translation engine. It enables a generalized translation tool for all kinds of applications, including text, voice, and full documents—including formatting.
- Custom data is data fed to an MT engine to build a specialization in a subject matter area like engineering or any other discipline with its own terminology.
Generic machine translation engines
The biggest machine translation providers have all moved towards NMT—each in its own way. On one hand, there’s the approach to reach a more general audience through a free, user-friendly tool. On the other hand, some engines also make it possible to tailor the tool to more specific business needs. Let’s have a look at the most popular, general-purpose MT engines:
Google Translate is considered one of the leading MT engines, based on usage, number of languages, and integration with search. Google Translate was one of the first mainstream providers to adopt NMT. The accuracy of Google Translate has been a point of both interest and contention for business owners and language industry experts alike.
Amazon Translate is also neural-based and closely integrated with Amazon Web Services (AWS). It may not be unexpected for many that Amazon Translate has accomplished impressive results in just a short time since its launch in 2017, given the power of its parent company.
Another cloud-based neural engine, Microsoft Translator is closely integrated with MS Office and other Microsoft products, providing instant access to translation capabilities within a document or other software.
DeepL is the product of a German-based company that is exclusively devoted to the development of a machine translation engine. It claims to have a more nuanced and natural output based on its proprietary neural AI.
Systran was the first company ever to offer machine translation for commercial purposes. Founded in 1968, it keeps following the latest technologies and introducing some interesting innovations itself—the latest being pure neural machine translation (PNMT).
Custom machine translation engines
Custom machine translation engines are trained to perform better for specific content types—also known as domains, e.g., technical or legal translations—or according to specific company guidelines. Key to this is relevant and high-quality training data in the respective domain, which can be used to “teach” the MT engine to perform similar translations for that specific use case in the future. One of the simplest ways to customize MT engines is using machine translation glossaries.
Google AutoML and Microsoft Custom Translator are 2 commonly used solutions for custom machine translation.
If implemented well, custom MT can deliver output with notably higher quality than generic MT. Nevertheless, machine translation customization requires a certain skill and effort. Fully customizing an MT engine can be a complex task, and each customization will be unique.
Interactive MT report: Uncover top performers
Find out how leading machine translation engines perform for different content types using the latest data in our quarterly machine translation report.
Benefits of machine translation
Before neural learning was introduced, MT was still very much a niche product generating translations of varying quality—sometimes bordering on unreadable or humorous. Modern machine translation tools have largely changed all of that and are increasingly becoming indispensable in business translation.
Increased speed and volume
With ongoing improvements in machine learning algorithms and hardware technology, MT is becoming even faster and more efficient. Not only is it capable of translating millions of words almost instantaneously, it’s also continually improving as more content is translated.
For high-volume projects, MT not handles volume at speed, it can also integrate with other software platforms such as content or translation management systems to organize that content. This makes it possible to retain organization and context as the content is translated into multiple languages.
With the major MT providers offering up to 100 languages—and some of them even more—translations can be delivered simultaneously to multiple target markets. This creates a win-win situation for both businesses and customers.
By eliminating language barriers and improving the customer experience, MT has boosted the accessibility of content, products, and services for potential buyers worldwide. At the same time, by reaching a wider audience, businesses can significantly expand their market share and improve their bottom line.
The combination of high-speed throughput and the ability to select from existing language pairs covering dozens of combinations means the use of machine translation services can cut costs and time for delivery, even when human translators are still post-editing the work.
Basically, MT does the initial heavy lifting by providing basic but useful translations. Human translators then refine these basic versions to more closely reflect the original intent of the content and ensure proper localization per region.
A look at the limitations of machine translation
The relatively low cost and minimal latency make MT an attractive option for businesses looking to efficiently expand their global reach. However, like any technology, it’s important to recognize that MT comes with its set of limitations.
Most issues can be effectively managed through the integration of human expertise—for example, machine translation post-editing (MTPE)—and increasingly through customizing your own machine translation engine.
Both approaches can help strike a balance between efficiency and linguistic precision, enabling businesses to leverage the strengths of MT while addressing its inherent constraints. Let’s take a look at some common issues related to MT.
|Issue to be aware of
|Accuracy and domain specificity
|By integrating machine translation glossaries, your custom MT model can adapt to evolving language patterns and contexts that matter to your business. Fine-tuning with human feedback also allows it to learn from human translators, correct errors, and improve translation quality over time.
|Incorporate cultural knowledge and context-aware algorithms to help your custom MT model understand nuances, idioms, and cultural references in the source language and consider the surrounding text to choose appropriate translations, capturing the intended meaning more effectively.
|Post-editing is one of the most effective methods to guarantee translation quality in low-resource languages. As your specific requirements evolve, you can explore a range of MT solutions—from learning models and data annotation to community involvement and open-access tools.
|While post-editing and human feedback are crucial for mitigating bias in MT output from generic engines, the optimal long-term solution is to audit and retrain your own custom MT model using diverse, unbiased datasets that don’t favor any particular group, perspective, or demographic.
|Machine translation post-editing
|Training linguists in MTPE is crucial for unlocking its full potential, as they gain insights into translation quality, accuracy, and cultural sensitivity. Ensure your translation vendor is ISO 18587:2017-certified and conduct periodical quality checks of post-edited content to enhance productivity.
|Combine MT with your translation memory. Once machine-translated output has been post-edited and approved, make sure to save these translations in your translation memory so linguists can reuse them in upcoming translation projects for improved consistency.
Implementing machine translation into existing translation workflows
As mentioned previously, the low cost and lack of latency of MT are compelling reasons for many growing businesses to include machine-translated content in the automation of translation and localization workflows.
Implementing an effective machine translation strategy doesn’t have to be a daunting task when you rely on cutting-edge cloud translation technology, allowing for automation and streamlined translation management.
For example, translation management systems have built-in settings to automatically run translation and send the output off as part of the human translator hand-off material.
By following a few essential steps and best practices, combined with a reliable translation management system, you can pave the way for a seamless enterprise machine translation process in the long run.
|10 best practices for implementing machine translation
Define your goals, objectives, and expectations from employing machine translation in your globalization program.
Audit your existing content and choose the right types of content for machine translation.
Determine your languages pairs—different MT engines are more suitable for certain language pairs than others.
Develop a timeline and financial plan—how much money and time you can spend on MT will determine how much you can accomplish.
Train your preferred MT engine with your language data if possible to increase the output quality in the long run.
If you go for MT post-editing, it’s important to ensure that your in-house translators or LSP are either trained in post-editing or at least open to the idea.
Agree on a pricing model for MT post-editing and be sure to involve all stakeholders, including your LSP, in the decision-making process.
Run samples before deployment to get an idea of the quality of machine-translated output or identify areas for improvement.
Deploy and keep improving: Bear in mind that the initial results may not meet your expectations, but the quality of the output will improve over time.
Machine translation vs human translation: Striking the right balance for the right use case
As machine language translation evolves, the decision between utilizing machine or human translation at the outset of a localization project is becoming less relevant. More and more businesses, language service providers (LSP), and translators are recognizing the benefits of machine translation post-editing (MTPE), which involves human linguists editing machine-translated content.
Machine translation post-editing is now largely considered a viable alternative to translating text from scratch.
When deciding on the appropriate translation method, it’s helpful to consider your content type and language pairs. Generally speaking, MT is well-suited for more structured content such as technical, legal, and intellectual property documentation, as well as internal communications. In contrast, marketing materials and other customer-facing content would benefit from a more human touch.
Let’s explore 3 key content types to determine the most effective translation method for the best results.
Use raw machine translation for low-impact and unambiguous content
Machine translation is said to be “raw” when the output doesn’t undergo human revision. In general terms, it’s best to keep raw machine translation for anything that won’t make or break your brand:
- Low-visibility or low-traffic content, such as internal documentation, website footers, social media posts for sentiment analysis, etc.
- Repetitive technical content that doesn’t need to be 100% accurate, just actionable, like instruction manuals
- User-generated content like product reviews, for which consumers generally expect low quality
- Quickly perishable content, like chat or email support messages, customer inquiries, etc.
- Large bulks of content with a short turn-around, such as hundreds of product descriptions that need to go live quickly
- Frequently amended content like feature and information updates
If you decide in favor of raw machine translation, it’s vital to ensure that you use the best-performing machine translation engine for your language pair and content. This requires significant testing or the use of an integrated auto-selection functionality.
Apply light or full post-editing to more sensitive content
For quality purposes, some content types and situations require post-editing of machine translation output by a human translator. This editing can be either light (LPE) or full (FPE).
Good news: You can aid the work of post-editors with traditional translation technology such as glossaries, termbases, and translation memories, as well as brand books and style guides. This will keep the brand voice and key messaging consistent across cultures and languages and is very feasible with MTPE.
Modern translation technology also makes it possible to identify and estimate the quality of machine translation output to concentrate post-editing resources where they’re most needed. Nevertheless, as a general guideline, the below cases require MTPE:
- Product titles: They are highly informative and concise, they tend to contain proper names and polysemous words, and their word order is usually relatively free, which can cause ambiguity.
- Translations between language pairs of dissimilar syntax, like Japanese and Spanish, because the reordering of words and phrases to well-formed sentences becomes more challenging for machine translation engines.
- Product descriptions: They need to be well-crafted and clearly state the product’s features or benefits without room for ambiguity.
- Content of medium visibility that needs to be as accurate as possible: knowledge base, FAQs, alerts, etc.
- Back-end meta information such as image alt texts and captions: While their visibility is low, a human needs to ensure that the target-language keywords are present.
Stick to human translation when branding and culture come into play
Brand-sensitive, high-traffic, and durable assets are best left in the hands of human experts. In other words, it’s generally advisable to avoid machine translation when the goal is to engage, entertain, or reassure the audience.
In such cases, a more human touch is your best bet, meaning a human translator will need to recreate the message in the target language in a non-literal way—you may have heard of this as “transcreation.” It’s the case of:
- Advertising landing pages
- Blog posts
- Newsletter campaigns
- Press releases
- SEO content
- Print advertising, etc.
10 key steps to creating a machine translation strategy
Learn how to design a machine translation strategy that can help your brand connect with international customers at full speed.
What makes the best machine translation software?
Choosing the best tool for machine translation can be complex as both general-purpose and specialized MT engines have unique advantages and limitations.
That’s why, regardless of whether you own the MT process or rely on an external machine translation service, you should ideally have access to different MT engines in one place to make the most of machine translation.
Utilizing a translation management system as the source of truth for all translation activities enables 3 key elements of successful MT deployment:
- Automatically selecting the optimal MT engine for your content type
- Quality estimation for improved post-editing efficiency
- Tracking key metrics to optimize productivity, turnaround times, and cost savings
Selecting the optimal MT engine for your content type—automatically
Different translation projects may require different levels and types of sophistication. This is one of the reasons why utilizing MT as part of a translation management system is so beneficial.
To get the best out of MT, you need to be able to assign the right engine for each type of content—and the most robust translation management systems have plugins or application programming interfaces (API) that connect them to different MT engines.
The most advanced systems even offer the ability to automate the selection process based on artificial intelligence or algorithms that scan the content and match it to the optimal MT engine.
Quality estimation for improved post-editing efficiency
An essential part of successful MT implementation is knowing where to direct post-editing efforts. When you’re able to measure the quality of MT output automatically, you can focus on the right segments instead of wasting time and resources where raw output is already of good quality.
This removes the guesswork from MT and improves post-editing efficiency, and it constitutes yet another reason why using MT integrated into a TMS is advantageous: The most sophisticated systems include automatic machine translation quality estimation capabilities that can identify which segments need more attention than others.
Tracking key metrics to optimize productivity, turnaround times, and cost savings
As mentioned above, MT is attractive to many organizations because it offers the potential for increased productivity, faster turnaround times, and ultimately, cost savings. However, not all MT engines deliver on this promise equally, so having the means to compare how different engines can affect the process is key.
A strong TMS lets you track the time and expenses of any translation project where MT is applied. With multiple MT engines in use, these metrics can be a strong indicator of an engine’s value: Is it increasing or slowing translator productivity? Does it indicate improved efficiencies over other engines over time? The answers to these questions will give you a better sense of its capabilities.
Make the most of machine translation with Phrase TMS
When businesses need to utilize machine translation at scale, they need technology that can provide them with the best of both worlds: efficiency and quality.
Phrase TMS, the enterprise-ready translation management system within the Phrase Localization Suite, makes it effortless for growing companies to leverage machine translation. Organizations achieve an unprecedented ability to enter new markets more quickly and efficiently.
Phrase TMS users can employ a dedicated machine translation add-on, Phrase Language AI, to implement MT into their translation workflow with fast and cost-effective translations that don’t compromise on quality.
Fully embedded into Phrase TMS, the advanced MT management features that come with Phrase Language AI let you:
- Start translating immediately with no developer time or effort using fully managed MT engines from leading providers like Google, Amazon, DeepL, or Microsoft.
- Add any of the 30+ supported generic and custom engines manually if you ever prefer to use a specific MT engine.
- Extend high-quality MT to every employee: The powerful Phrase Language AI API allows you to scale the value of MT with company-wide access for to secure company-approved machine translation.
- Enjoy unlimited machine translation for post-editing workflows so linguists can work more efficiently.
- Work with the best engine—auto-selected, based on your language pair and content type.
- Automatically filter out content that shouldn’t be machine-translated.
- Delegate quality testing, legal and security evaluation, setup, and payment of machine translation engines to dedicated machine translation experts on the Phrase team.
- Leverage your translation memories to increase translation quality by up to 50% with Phrase NextMT—the first TMS-ready MT engine.
- Ensure the MT engines use your preferred terminology with the correct morphological inflection—reducing post-editing effort.
- Preserve formatting and placeholder tags from source to target content automatically.
- Get a score for each machine-translated segment, based on past performance data, to post-edit only where needed.
- Achieve up to 55% cost savings with MTPE compared to human translation.
Discover advanced MT management features within Phrase TMS and push MT technology to the next level with our enterprise-ready solution.
Speak with an expert
Want to learn how our solutions can help you unlock global opportunity? We’d be happy to show you around the Phrase Localization Suite and answer any questions you may have.
Last updated on October 11, 2023.