There are more machine translation platforms than ever available for businesses that want to manage their own localisation projects. But which is the right for your business? And can any of them replace a professional Language Services Provider? We review 5 machine translation platforms to help you make that choice.
Amazon Translate uses Amazon’s neural machine translation engine to deliver “fast, high-quality, and affordable” translation services to businesses. Use cases include translating websites and apps, translating large volumes of text and multilingual customer service.
Top Language Pairs
According to Intento’s 2020 report on commercial machine translation engines, Amazon Translate’s best language pairs include English and Portuguese, English and Korean, English and Spanish, English and Russian, English and Finnish, English and Japanese, and English and Dutch.
However, the industry/sector and the type of source content also influence translation quality.
Amazon Translate is customizable via its Custom Terminology and Active Custom Translation features.
Price: Approximately $15 per 1M symbols
DeepL calls itself the “word’s best machine translation”, with translations that are preferred by professional translators. They have a focus on data security, encrypting connections to their servers and deleting texts immediately after translating. DeepL also provides the option of translating whole documents while retaining the original fonts and formatting (although the original formatting may still need to be tweaked to accommodate text expansion and contraction).
Top Language Pairs
DeepL has a limited number of languages and language pairs available compared to other providers: only 7 languages and 42 language combinations. With a focus on quality over quantity, it excels in English to German translations, according to Intento’s evaluation.
Choose the tone of your translation (formal or informal) and create your own glossary of approved translations for certain language pairs.
Price: DeepL does have a limited free option, but if your translation needs exceed the free version or you need access to their API, you may end up paying more than you would with some of the other providers, somewhere in the range of $20 per 1M symbols.
Microsoft Translator offers businesses a number of services including text translations for apps, websites documents and more. They also offer an end-to-end, real-time speech translation service.
With that said, their overall translation quality was ranked lower than competitors in Intento’s evaluation.
Top Language Pairs:
Microsoft Translator supports more than 70 languages. However, they didn’t make it into Intento’s list of best MT engines for any language pair. With that said, Phrase.com did rate them highly for English to French translations, based on the preferences of their in-house translators.
Microsoft Translator makes it relatively simple to use previously translated documents to “train” their NMT engine to produce translations customised to your business and industry.
Pricing: $10 per 1M characters, or less, depending on volume.
Google Cloud Translation
Google offers free translation via Google Translate, but they also offer a paid machine translation platform for businesses. Choose from the Basic Translation API, the Advanced Translation API, or use AutoML to train your own custom machine-learning models.
Top Language Pairs
Google offers 100+ languages, but their top-performing language pairs include English to French, English to German and English to Russian.
Customers can use previously translated texts to “train” a customised model using AutoML translation and add a glossary with the Advanced Translation API.
Pricing: Google Translate may be free, but Google Cloud Translation costs around $20 per 1M symbols.
Systran Pure Neural MT
Systran makes big claims for its NMT engine, specifically that it can produce a translation “overachieving the current state of the art and better than a non-native speaker.” (A quick reminder: the linguists we work with ARE native speakers because that is the gold standard for human-powered translation.) There are 55 languages and 140 language pairs included in their professional products.
Systran offers limited free text translation, but the paid options (Systran Translate Pro and Systran Pure Neural Server) offer more: text translation, document, translation, webpage translation, and audio file translation as well as add-ons for translating websites as you browse and for translating within various Microsoft Office products.
Top Language Pairs
Systran PNMT’s top language pairs include English to Portuguese, English to Turkish, English to Romanian, English to Arabic, and English to Polish.
Systran’s paid plans allow businesses to customise based on industry as well as to use translation memories and user dictionaries.
Pricing: Generally speaking, Systran is amongst the cheaper options, with pricing that averages around $10.50 per 1M symbols.
Why Use an LSP Instead?
With so many companies marketing machine translation platforms directly to businesses, why and when does it make sense to use an LSP instead? Unless you already have a dedicated internal team with all of the skillsets needed to effectively manage translation and localisation from start to finish, you’ll see better results with an LSP for all but the most low-stakes projects.
Machine translation’s accuracy is variable, and even with customisation, in most cases, you need a human post-editor to examine the content and check for mistranslations and awkward phrasing. If you’re translating into more than one language, consider that the performance of the top MT platforms varies significantly depending on the languages tested. In fact, Intento’s report on the state of the machine translation industry found that “models from eight different vendors were required to get the best quality across popular language pairs.” The bottom line: If you’re using MT and accuracy matters, a robust quality-control process is essential.
There’s also the issue of making sure that your translated content sends the message you want to send. Cultural differences mean that it’s often necessary to make changes to the original content to preserve the original impact of a message across different languages.
Neglecting these other aspects of successful translation and localisation can result in embarrassment and reputational damage. For example, consider Amazon’s roll-out of their own Swedish-language website. Despite Amazon offering one of the top-rated NMT engines, their initial Swedish-language launch was marked by gaffes, including NSFW translations of product descriptions and accidentally using the Argentinian flag in place of the Swedish flag.
The Advantage of MT with Added Value
Quality LSPs offer more than just words- they are trusted partners ensuring that your brand messaging doesn’t get lost in translation. With a team of expert linguists, designers, multilingual copywriters, compliance experts, consultants and project managers to coordinate it all, working with an LSP like K International saves you time, protects your business and gives you peace of mind, too.
At K International, we know when and how to use MT to save our clients time and money, what projects require more hands-on human involvement, and how to design a quality assurance process that prevents costly and/or embarrassing mistakes. Contact us for more information on how we can help you with your next translation or localisation project!