
Translation management
Machine translation
The promise to enable efficient and accurate communication across languages has been a driving force behind the development of machine translation (MT). What began as an experimental endeavor back in the 1950s—translation was one of the first applications of computing power—became a viable productivity tool in the 21st century.
Today, AI-driven machine translation tools are revolutionizing global business operations. One of the leading MT providers is DeepL, a neural machine translation (NMT) technology provider based in Germany. In this guide, we will explore how DeepL works, its pros and cons, and best practices for utilizing it in professional translation projects.
DeepL was founded in 2009 in Germany as Linguee, an online dictionary, which set out to create a neural machine translation system that could produce translations of a much higher quality than traditional statistical machine translation (SMT).
The engineers at DeepL applied the newest deep learning technique—hence the company’s name—to train the models on the existing data on Linguee’s database.
Since 2017, DeepL has become extremely popular—more than a billion people have used its services to date. It provides support for 31 languages, with 650+ possible translation combinations.
Users can choose between the free and paid versions of DeepL, and between the web interface and the standalone translator. The free version is suitable for personal use, while the paid version offers more features for businesses.
DeepL is no longer just a translation service, though. It has now also had a go at artificial intelligence with a focus on text generation. Its newest offering, DeepL Write, launched in early 2023, seeks to become an English writing assistant and outperform rivals like Grammarly.
Free download
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.
Not all machine translation is created equal, and over time, systems have become more and more sophisticated. Until 2016, MT systems were either rule-based—which relied on numerous manually crafted rules—or statistical—which translated based on multilingual corpora (large bodies of parallel texts), in a word-based or phrase-based fashion, looking for statistical patterns.
Nowadays, most mainstream MT systems use neural networks. This is called neural machine translation, a form of end-to-end learning where the program’s neural network takes into account the whole input sentence at each step when generating the output sentence, rather than a few words on either side of the translated term.
DeepL is an example of an NMT system—thanks to deep learning algorithms, it produces translations that are more human-like than those generated by statistical MT engines. Its network architecture also enables it to learn from vast amounts of data and adapt to new contexts.
As mentioned previously, DeepL’s core strength lies in its NMT system, enabling it to produce more accurate and natural-sounding translations compared to traditional statistical MT methods. Let’s take a look at some of its most prominent capabilities:
The universe of machine translation tools is vast and varied. Aside from DeepL, some of the most well-known MT providers include:
Launched in 2006, Google Translate switched from a statistical model to an NMT one in 2016—a few months before DeepL launched.
It supports over 133 languages and is free to use (except for API usage to translate your own website or if you go over 500,000 characters per month). Its highly accessible interface and intuitive design make it a popular choice for casual users. Moreover, it can translate entire websites, images, and speech.
Founded in 1968, Systran was the first commercial MT software on the market. Systran is also the only company with an open-source ecosystem for neural machine translation and neural sequence learning: OpenNMT.
Systran Translate supports over 50 languages, and users can add their own glossaries, dictionaries, and corpora to personalize the output.
Integrated into Bing (Microsoft’s search engine) and as a built-in function in Microsoft Office applications, Microsoft Translator launched in 2009 and is based on the newest neural network technology with an attention-based model.
Nowadays, it’s also available as a standalone mobile app for both iPhones and Android devices. It supports over 100 languages and enables speech and text translation.
One of the youngest players in the field, Amazon Translate was released in 2017. It uses a neural machine translation engine and has achieved impressive performance levels in its short lifetime.
Users need an AWS account to access the range of functionalities that Amazon translate offers—customization (terminology and parallel data), encoding terminology, and batch translation (Amazon S3), to mention but a few.
Tencent Machine Translation (TMT) is one of the newest entrants to the market. It currently supports 10 languages and combines both neural and statistical machine translation models.
Tencent Machine Translation is particularly notable for its Chinese-language capabilities, having obtained the highest recorded human evaluation scores for English-to-Chinese translation and the highest automated scores for Chinese-to-English translation.
Is DeepL better than Google Translate? The answer to this question, like anything else, depends on the context and the specific needs of the user. Generally speaking, DeepL is often seen as a more accurate MT engine than Google Translate.
That said, it does come with some limitations: DeepL’s language selection is more limited than Google’s, and to access its full range of features—like full document translation that retains the original formatting or the ability to choose between formal and informal register—you’ll need a premium account.
A feature that DeepL offers that users can’t get from Google Translate is the ability to click on any translated word in the output box to quickly view alternative translations. If you select a different translation from the one that DeepL has suggested, the rest of the text will automatically update to reflect your choice.
Security-wise, in the case of their free versions, both Google Translate and DeepL retain the history of the text you translate. DeepL Pro, by contrast, offers world-leading data protection standards and deletion of your texts immediately after translation. This makes it particularly suitable for sensitive content.
Overall, both providers offer useful solutions for machine translation. If you’re just looking for a cheap, easy-to-use translation tool with a wide range of language options that you don’t need to customize, then Google Translate may be the better option. On the other hand, if you need highly accurate translations that you can customize and need to be sure your data is protected, then DeepL is likely to be the better choice.
Establishing the accuracy of a machine translation engine can be notoriously challenging. The results vary depending on the language pairs involved, the types of texts you’re translating—a highly technical financial document isn’t the same as a blog post—and the level of customization that you need.
Moreover, because language is dynamic and factors like the author’s intended meaning can’t be measured, accuracy takes on a relative definition. For example, the expectations of grammar and punctuation accuracy for a novel will differ from those for a social media post.
That said, based on general sentiment around the web, users report that DeepL tends to be quite accurate, especially when it comes to European language pairs. DeepL’s own experiments back this up. However, because the way the company presents the data could potentially be seen as biased, a more reliable indicator of accuracy is user feedback—which tends to be positive, mainly in terms of:
At the end of the day, each user will need to make their own decision after trying how DeepL performs in their use cases. The good news is that the free version allows you to do just that without incurring any cost.
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.
Neural machine translation systems such as DeepL open up a whole world of possibilities. As one of the leading NMT technology vendors, DeepL can be:
That being said, even though DeepL’s development has reached great heights, hardly any machine translation engine will likely ever be perfect—and DeepL still has some way to go before catching up with human-level accuracy.
How much DeepL will continue to improve in the coming years remains to be seen and is difficult to predict. For now, it still falls short of human-level skills such as:
While the ability of machine translation engines to handle these types of tasks remains limited, manual post-editing is still likely to remain an essential part of any translation workflow in the near future.
When using DeepL for machine translation, it’s important to stick to the use cases where you can get the most out of it without compromising on quality. These include:
Most of these use cases will require light machine translation post-editing (MTPE) to ensure accuracy and clarity, and you may even get away with using the raw output if the content is not mission-critical.
Some other types of content that require a higher level of accuracy will require more time-intensive post-editing. Among them, we find:
Modern translation technology, such as localization platforms, has become an enabler of efficient global expansion efforts by revolutionizing the way enterprises manage, translate, and deliver multilingual content. These tools streamline the entire process from end to end.
Most of these solutions also integrate machine translation engines, allowing users to quickly convert large amounts of text into the target language within the same interface they already use to pull in, translate, review, and export content back to the system of origin. The gains in productivity alone are significant and, when combined with the cost savings associated with leveraging MT in the first place, the resulting ROI can be spectacular.
By contrast, when you use MT as a standalone engine that doesn’t live in the same system as your content, you’re effectively adding an extra layer of complexity to the translation process—just think of the hassle of manually uploading files or downloading results. The resulting bottlenecks, disjointed workflows, and lack of visibility into the entire process inevitably affect the delivery time and accuracy of your translations.
Take the case of DeepL as a fully managed MT engine in an enterprise translation management system such as Phrase TMS. The system automatically uses the best machine translation engine for each job, and its AI-based solution filters out content that shouldn’t be translated:
It’s clear that DeepL has come a long way since its humble beginnings, and today it can be used to quickly generate high-quality translations for a variety of use cases.
However, to unlock its true potential, you need to use it for the right kind of content, with differing levels of post-editing according to the use case, and of course, within the right technology. A translation management system such as Phrase TMS is the perfect context for this purpose, allowing you to access all the features that make DeepL shine.
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.
Last updated on September 24, 2023.