Complete guide to machine translation tools in 2024

machine-translation-guide

Machine translators continue to be widely adopted in 2024. MTPE service is becoming more and more popular. These automated translators are getting more accurate by the year, converting large volumes of text into other languages in seconds. This allows the human translator to save time and focus on more complex nuances such as jokes, puns or cultural references.

There are a multitude of online machine translators to choose from and deciding on the most suitable one can be a challenge. This article explores the advantages and disadvantages of five well-known machine translators, helping you find the most suitable option for your personal needs.

The most popular machine translation tools in 2024

LanguagesFile uploadAPIFree versionSpecial featuresSuitable for
DeepL32YesYesYesAI-powered text editing for English and GermanCasual users and small businesses
Google133YesIn paid versionYesExtensive image translationCasual users, small- and medium-sized businesses
Microsoft135In paid versionIn paid versionYesSpecial functions for educational and business settingsSmall- to enterprise-level businesses
Amazon75YesIn paid versionFree trial for 12 monthsBrevity function, profanity maskingSmall- to enterprise-level businesses
ChatGPT∼100In paid versionIn paid versionYesCustomisable output stylesCasual users to enterprise level businesses
The most popular machine translation tools in 2024

For this purpose, we analyse a sample paragraph which was originally in Swedish, and translated into English by the machine translators of DeepL, Google, Microsoft, Amazon and ChatGPT. The paragraph is sourced from the website Swish.nu, Sweden’s leading mobile payment application.

DeepL

DeepL allows you to translate up to 5000 characters per month for free, while premium plans starting at €7.49 per month offer unlimited text translation. 

In the free version, users are allowed three non-editable file translations each month, and the file size for uploads is restricted to 5 MB. Additionally, users can create one glossary containing up to 10 entries.

The paid plans include up to 100 editable file translations, supporting file uploads of up to 30 MB. Additionally, users can manage up to 2,000 glossaries with 5,000 entries each. The paid version also provides advanced functionalities like team administration, single sign-on (SSO), and seamless CAT tool integration. In addition, DeepL’s paid versions focus on maximum data security.

Pro versions allow you to specify whether the translation will be formal or informal. Furthermore, when translating into Japanese, you have the option to select either an informal or a polite tone, a significant linguistic distinction.

DeepL also has a mobile app, allowing on-the-go translations and image translation. On the other hand, DeepL image translation only supports a few languages and cannot read Greek or Cyrillic characters.

Users commend DeepL for its high level of accuracy. However, it supports only 31 languages, a limitation compared to Google Translate’s 133 languages.

With a simple and easy-to-follow sentence structure, DeepL machine translation offers a better flow than most other programs. In the first two sentences, DeepL has eliminated the Swedish med (‘with’) and på så sätt (‘in this way’), which are redundant in English. In the third sentence, DeepL has steered clear of verbatim translation of the source text; While the Swedish version says butiker med för mycket folk i (literally ‘stores/shops with too many people in them’), DeepL has used ‘crowded stores’.

On the other hand, DeepL has translated ombemannat as ‘unmanned’whereas in this context it means ‘remanned’ or ‘restaffed’. This suggests that DeepL excels in concise and well-flowing translations, but may not be the optimal choice for achieving a precise word-for-word translation.

Google Translate

Although often seen as the epitome of incorrect translations, Google Translate has come a long way since its switch to neural machine translation.

Neural machine translation (NMT) is the newest approach to machine translation. Although NMT also uses language corpora and statistical calculations, neural machine translation engines are based on artificial neural networks. One could say that artificial neural networks are trained to translate. They are programmed to search for patterns from the texts provided (somewhat similarly to the human brain) and then provide translations themselves based on the patterns they detect. Unlike statistical translation systems, NMT engines look at the entire source sentence instead of individual phrases, providing more grammatically correct and fluent translated sentences.

Google Translate offers versatile features like app integration and a website translation widget, and supports 133 languages. Its offline mode and batch translation for files enhance usability.

One notable strength is Google Translate’s image translation capability, covering an extensive set of languages and alphabets. The online translator can translate images so that the font and layout remain the same, making it really efficient at fast translation of advertisements or invitations.

However, using Google Translate’s free online version for confidential information is not recommended, as the data can be processed by Google’s algorithms and potentially utilised for various purposes, including marketing decisions. Higher levels of data privacy are provided by the paid version, Google Cloud Translation API.

Unlike DeepL, Google has used the sentence constructions ‘With store apps, the customer can…’ and ‘In this way, the queues disappear’. While not actually wrong as such, it makes the text longer and less easy to read. 

Google Translate has also made a translation mistake. In the original text, 18 till 29-åringar har lämnat butiken för att kön var för lång translates to ‘18- to 29-year-olds have left the store because the queue was too long’, but Google has translated it to ‘18 to 29-year-olds have left the store for being too tall’. This is a serious translation error that completely changes the meaning of the text.

This said, Google Translate correctly translated ombemannat to ‘restaffed’, which DeepL got wrong. According to Swish the correct term would be ‘remanned’ but if the consistency in terms is not essential, ‘restaffed’ conveys the meaning very well.

Microsoft Translator

Microsoft Translator, with Bing Translator as its web interface, is Microsoft’s translation service. It shares similarities with Google Translate, but Microsoft Translator offers a more cost-effective price-to-characters ratio. Microsoft Translator is twice as affordable as Google Translate. It provides a free monthly translation limit of 2,000,000 characters, with additional characters incurring a cost of $10 per million.

In contrast, Google Translate allows free translation of 500,000 characters per month, and any surplus characters cost $20 per million characters.

For casual users, Microsoft Translator stands out by suggesting commonly used phrases across various categories, including occasions, romance, and emergencies. Its mobile app excels at image translation and real-time speech recognition. With 105 languages and 12 speech translation systems, Microsoft Translator offers broad language coverage.

For professional users, Microsoft Translator’s features include app integration, a website translation widget, and a project management dashboard. It offers different specialised features depending on whether it is meant for personal, business, or educational use.

Microsoft Translator users have reported the most problems with the accuracy of the translation, claiming that it particularly struggles with longer sentences. However, its integration into the Microsoft ecosystem, including Office, Skype, and Bing, makes Microsoft Translator the first choice for many.

Much like Google Translate, Microsoft Translator retains longer sentence structures, potentially making the text less easy to grasp. Nevertheless, this translation shows no apparent mistakes.

Amazon Translate

To use Amazon Translate, you must first create an account on Amazon Web Services. The Amazon Translate Free Tier offers a complimentary service for up to 12 months, enabling users to translate up to 2 million characters per month.

Amazon Translate supports translation between 75 languages. It can be integrated into various applications, websites, or services.

Emphasising data security, Amazon Translate incorporates encryption, access control, and compliance with industry standards to safeguard user data.

Amazon Translate also offers a range of additional settings. Users can create custom terminology, ensuring precision in domain-specific translations. It is also possible to opt for a brevity function, making it suitable for texts with specific length requirements, such as captions, subtitles, headlines, or form fields. Moreover, users can choose the formality of the translation output and utilise profanity masking to filter and manage inappropriate language.

The output from Amazon Translate closely mirrors that of Microsoft Translator. While it’s longer than DeepL’s result, it doesn’t contain any translation errors.

ChatGPT

Although ChatGPT is not created specifically for translation, it can still produce great results. ChatGPT accommodates around 100 languages and casual users can freely access the basic version with unlimited interactions. For those seeking advanced functionalities, GPT-4 starts at $20, offering features such as file translation, API access, and glossary creation.

In contrast to the other translation programs mentioned in this post, ChatGPT allows users to provide prompts for adjusting formality, length, style, and output type. For instance, if you desire a more casual and humorous tone for your website’s translation, you can say that to ChatGPT. Alternatively, if your text needs to take on a formal letter style, you can instruct ChatGPT to tailor it accordingly.

However, a notable downside of ChatGPT is the absence of updates since January 2022, which makes it less adaptable for rapid language changes and improvements. The process of creating prompts for accurate translations can be time-consuming.

This is a translation generated using only the ‘translate’ prompt:

ChatGPT’s translated output closely resembles that of Google, Microsoft, and Amazon. However, a grammatical mistake is noticeable in ChatGPT’s first sentence, where ‘customer’ is singular, but ‘mobile phones’ is plural. This indicates that, regardless of the machine translator’s reliability, a human translator should always perform a thorough grammar check.

Below is the rephrased translation, using the prompt ‘write in a humorous and down-to-earth manner’:

The Chat GPT prompt rewrote this text using entertaining metaphors, creative adjectives, intriguing questions and target group-related elements.

If the result of the prompt is not exactly what you want, you can always change it. Nevertheless, as mentioned earlier, this process can ultimately be time-consuming.

In summary, the distinctions between the machine translators available online are minimal. Each translator possesses unique strengths and weaknesses, which are often unpredictable and context-dependent. Therefore, proficiency in both the source and target languages is essential for the user of machine translation.

Examining the particular paragraph, DeepL demonstrated a more independent approach, omitting redundant sentence constructions in English. In contrast, Amazon and Microsoft proved to be the most dependable, displaying no noticeable translation errors. It is crucial to note that results can fluctuate depending on diverse input texts, and even the same machine translator may yield different outcomes for the same text on different days.

GET IN TOUCH, WE’D LOVE TO HEAR FROM YOU.

Don’t forget to add your e-mail address and files and be sure to submit your enquiry. We will be waiting.