It's hard to imagine our globalized world without machine translation engines. Technologies are changing, developing, and improving daily; machine translation is no exception. They started as rule-based and statistical engines and now grew to use neural networks. This article will cover some of the most well-known machine translation engines. Check out our overview and pick your preference.
One of the most used machine translation services, Google Translate, covers over 130 languages. Based on the statistical method of machine translation in 2006, Google Translate became a neural network-based MT in 2016. This tool is pretty simple to use in both web and application versions, and as a result, it reaches over 1 billion app installations and translates 100 billion words per day.
In 2017 Google Brain team made a huge step for the machine translation industry by releasing the Transformer (architecture of deep neural networks), which by the way, uses for wide-knowing ChatGPT. The main peculiarity of the transformers is the so-called "attention mechanism" that allows parallelized text processing more easily than RNNs and can be trained faster.
Google Translate is still inaccurate if we talk about understanding the context, but the quality can differ based on the language pairs. Thus the English-Spanish pair has over 90% accuracy, while English-Armenian is only 55%. Despite these downsides, Google Translate is one of the best solutions on the market, with various languages, free access, and API.
Note: Lingohub uses Google Translate power in combination with DeepL and Amazon Translate to provide effective machine translation for customers.
Microsoft Translator (launched in 2009) is a part of Microsoft Cognitive Services and as well as Google Translate integrated into Google, related to the Microsoft search engine - Bing. It supports over 100 languages and proceeds to extend the language base. In 2022 Microsoft started integrating an entirely new AI technology Z-Code into their translating engine (Z-code is a part of XYZ initiative for combining AI models for text, visual, and language.)
The blind test made by the Microsoft team showed an average translation improvement of 4%. Some language pairs have a much better result, for example, English-Slovenian - by 15%, Japanese to English by 7.6%, and English to Arabic by 9.3%. Since it's an open cloud API, Microsoft Translator is easily integrated cross-platform. It's available both free and under a subscription. However, the API for business applications requires a subscription plan. The integration with other Microsoft products also allows translating Microsoft Office documents, Skype messages, and Linkedin posts between all the provided languages.
We previously wrote about DeepL on the Lingohub blog, but let's take a closer look by comparing it with other MT systems in this article. DeepL is a worthy opponent to industry giants like Google Translate and Microsoft Translator, and in this article, we will describe why. One of the "youngest" solutions from our list (DeepL was launched in 2017) positions itself as "The world's most accurate translator."
DeepL supports only 29 languages and 800 combinations, unlike the previous two solutions. Service uses the algorithm with convolutional neural networks (CNNs) and is trained by the most extensive world database of human translations - Linguee. Linguee is crawling the Internet for bilingual texts. Then the algorithm and company linguists evaluate the quality of the translations and then provide proofed results for users. And here, the DeepL shines.
DeepL is opting out of a wide scope of offered languages for translation. Instead, they are focusing on a small number of efficient translations of high quality. This strategy is the winning one. More natural translations cut costs, especially for those who usually choose to invest in quality checks.
In the case of accuracy, the DeepL team did their research. DeepL, Amazon, Google Translate, and Microsoft Translate translated 119 paragraphs. External professionals reviewed the translations in the blind method (they did not know which tool translated the text.)
Based on the graphic, DeepL is much more accurate for the next language pairs:
- English → German
- German → English
- English → French
- French → English
- English → Spanish
- Spanish → English
DeepL Translator is free, while DeepL Pro, a business-oriented version, offers the API for a fee. We're excited to see what will be their next step of development.
As well as a DeepL, Amazon Translate is a relatively new solution launched in 2017 by Amazon Web Services (AWS.) It is a neural machine translation service that offers fast and accurate text and speech translations. Currently, Amazon Translate supports 75 languages and dialects. The service is constantly expanding the language base and adding new languages. "Inside" Amazon Translate is a "deep learning" method for neural networks.
This service provides an exciting feature - Active Custom Translation (ACT) - which means businesses with specific vocabulary can train the system independently. It is easy, like apple pie - the user must only upload the TMX, TSV, or CSV files to the system to "teach" it. Users can integrate Amazon Translate via API to their solution and manage translation results.
Last but not least, Apertium is an open-source machine translation platform available on GitHub . Unlike all the examples from this article, it is rule-based software. It offers translation for 45 languages.
It started as a project developing in Spain, with several Universities and the government involved. This is obvious in the very translator - several specific Spanish dialects were targeted for the very beginning. The variety of dialects and versatility of an engine like Apertium might give it a cutting edge in the future. We will make sure to follow its progress.
Ready to localize?
Lingohub already offers you powerful engines to speed up your localization process. You can use our machine translation feature that, as we mentioned, combines the DeepL accuracy and Google Translate and Amazon Translate language variability.
Save the time of your translators by providing high-quality and automated pre-translating. Book a short demo call for more information, or try Lingohub with a free 14-day trial.