How the trade-off triangle affects your machine translation choice

If you’re looking to get large amounts of text translated quickly or to get a quick understanding of a document, then machine translation (MT) can be a great way to do that – but it can be difficult to choose the right machine translation solution.

The trade-off triangle helps you understand how different machine translation methods compare with each other and what kind of translations they produce.

What is the trade-off triangle?

The trade-off triangle is a graphical representation of the three factors that drive machine translation choice: quality, speed and cost. It is also known as the ‘Pareto principle’ after an Italian economist who first described it in 1897.

The trade-off triangle has been a staple of MT research for many years, but its practical implications are often missed. Most MT evaluations use standardised test collections, which generate scores for one specific trade-off (normally speed), but the graphic simultaneously shows all three factors.

Machine translation and the trade-off triangle

When deciding on what machine translation solution to use in your business, there are several questions you should ask yourself. For example, am I looking for a low-cost solution? A quick turnaround? Or high quality?

These questions can leave you confused. To get the machine translation that best fits your needs, you should consider the trade-off triangle. It is a good way of visualising three opposing forces or elements. It is also referred to as the ‘triple constraints’ of project management.

If you happen to choose in favour of one element, the other two elements will be affected by your choice. You can use any two at a time, but not all three simultaneously. What are the priorities of your company? Your answer should guide your choice.

machine-translation-triangel

The image above illustrates the trade-off triangle for machine translation that determines your choice. The following trade-off effects should guide your decision.

Cheap and fast

This choice means developing something quickly and cheaply. An example here is seamless machine translation. In this case, secure artificial intelligence (AI) powered machine translation engines (learn more about Semantix Secure Machine Translation) are integrated within the local environment. They are accessed through a closed web-based interface. Translations are executed ‘en masse’ at the click of a button.

Through seamless machine translation, you get instant, searchable text-file translations. These are cheap to process and are ideal for first-pass review.

Fast and good

This involves the creation of higher-quality machine translation engines, while still leveraging the speed of machine translation. There is a lot of research and product development done during the project initiation phase.

One example is enhanced machine translation. It allows you to access key information in a foreign language at machine speeds but with the addition of keyword glossaries and industry dictionaries. While still as fast as regular machine translation, you can also train customised machine engines with glossaries and dictionaries that improve the overall quality and consistency of your translated materials by more closely matching your company’s vocabulary.

Enhanced machine translation allows you to process larger native file formats to improve readability and accurately match formatting.

Good and fast

This option produces the highest-quality results using machine translation. Understanding the following characteristics of machine translations is vital for companies and businesses.

One example is Post-Edited Machine Translation. Here, a human translator revises machine-edited text. Post-editing the machine-translated text reduces the linguist’s workload by maximising the translation technology. Options can vary from light to full editing, depending on production requirements.

This machine translation process draws from the translation memory software. It captures company or matter-specific preferences and saves finalised translations for future reference and training. If you plan to publish your text, it is important to employ a human translator to post-edit it.

The take-away

In conclusion, the trade-off triangle is an important consideration when choosing an MT process, but it is not the only one. Other factors, such as turnaround time, requirements for human intervention and supported languages, can all play a part.

It is also important to remember that the trade-off triangle forces you to choose only two of the three options. Your choice should also be determined by your goal. Do you only need to understand what it says in the other language, or do you want to use the text to communicate with others? These kinds of questions should help to guide your decision.

Which one would you choose?

If you’re interested in learning more about the trade-off triangle for machine translation, or if you need help deciding which system to choose, don’t hesitate to get in touch with us. We would be happy to help.