How to Choose a Machine Translation Engine

As artificial intelligence advances, machine translation engines are also becoming more and more reliable. While many translation jobs are still more fit for human translators, a growing number of jobs are done either entirely by machine translation or through a mix between MT and human translation. If you are also considering using MT in your projects, here’s a guide for you on how to choose the right machine translation engine. 

Why use machine translation? 

Machine translation can be a perfect tool for translators as a supplement to their own knowledge and the toolset they use (such as translation memories, term bases, dictionaries, or any software used in the project). Besides speeding up your work, an MT engine can help you serve as a starting point for further human translation efforts (known as machine translation post-editing or MTPE). 

But MT is not only for the use of individual translators. LSPs and enterprises can also take advantage of enhancing their work with machine translation. 

Of course, there are many factors (such as the topic of the text to be translated), but generally, when you have large amounts of text, it is much more efficient to use a machine translation engine to go through all the data in a short time. The translation output usually requires some human editing or correction, but the use of MT undoubtedly makes the process smoother and quicker. 

By using machine translation in your workflow, you can reduce the translation turnaround time (and thus, the costs) by up to 50%. 

What to consider when choosing a machine translation engine 

Once you've decided to use MT in your translation project, you are now facing the question: How to choose the right machine translation engine for that specific task? 

Let’s see what to consider. 

Content type & quality 

First of all, you have to know what type of content you’re translating, and what level of quality you are looking to achieve.  

Some documents just need to be machine translated, without requiring any post-editing after the translation process—these documents are usually manuals or instructions, or sometimes internal papers or files where quality is not as important and translation only serves as a way to provide a basic understanding of these texts. 

Lastly, if you are translating more creative documents, such as literary texts, creative, or marketing concepts, machine translation might not be the best suited for that project—these types of texts require the wits and creative thinking that human translation can provide. 

The topic of your text 

Based on Intento’s 2021 report where they evaluated fifteen machine translation engines, it was stated that out of those fifteen engines, each one was best for something. Therefore, how you choose your machine translation engine depends heavily on the topic of the content you want to translate. 

According to Intento’s report, "the highest MT quality is available for Computer Software, Legal Services, and Telecommunications, with Software Strings and Documentation, Support Content, Policies, Processes and Procedures being the most accessible content types.” On the other hand, the most difficult texts for machine translation engines were documents from the Professional and Business Services sector, as well as Sales & Marketing Content. 

The more complicated a text is, the more beneficial it is to use neural machine translation (NMT) enginenes—as of now, our first choices in the premium translation market would be DeepL and SYSTRAN. This type of MT engine is best able to identify the subject of a text, and accurately predict what words and expressions to use in the translation output based on the subject matter of your source text. 

The language pair 

Naturally, the languages themselves are one of the most important aspects of translation. Many people are familiar with Google Translate or DeepL—but are these two all it takes? 

Again, in Intento’s report, among the fifteen MT engines that were evaluated, DeepL did prove to be in the lead for many languages paired with English (opposed to German or Spanish). However, if you are translating between English and Finnish or English and Japanese, your first choice should be ModernMT. 

If you’re translating from one source language into different target languages, your best bet is to combine the MT engines that work best with certain language pairs. If you are curious you can even try more than one engine on the same text to see which one is closest to the quality you want to achieve or which one requires the least post-editing. 

Turnaround time & budget 

Finally, we have arrived to one of the major selling points when it comes to machine translation: the price. Using MT can significantly reduce turnaround time, especially if your project features large bodies of text or excessive amounts of data. 

However, when looking at this aspect of translation, post-editing likely plays a big role in the translation process. When you want to have the most accurate translation of a large sum of text within a short time, human involvement is crucial. To help you with this, it’s best to consider the subject matter and language pair first to find the MT engine that provides a target text that requires the least editing. As always, some like relying on best practices, while others would rather experiment. 

Machine translation engines in memoQ 

At the moment, you can choose from more than twenty machine translation engines in memoQ. To learn which ones are available and how these integrations work with memoQ, visit our Machine Translation page. 

Summary 

Working with machine translation can really make a difference in a translation project. It can speed up work, save you money, and reduce turnaround time. However, in most cases, you cannot rely on machine translation only—you'll need a hybrid workflow of MT and human translation to achieve the quality that you need. Make sure you always choose the right machine translation engine and MT workflow for your project. 

Important note for freelance translators: Before you start working with a cloud-based machine translation engine within memoQ for a certain project, please make sure that your contract/NDA with your client does not prevent you from using it. 

What about you? Do you prefer translating everything yourself, or do you like using MT in your projects? What do you consider when you’re evaluating MT engines?