24/10/2024

Many companies approach machine translation (MT) like an Olympic competition: they think it only makes sense to use it if the machine is trained and customised to their needs, so they assume it will involve a huge amount of preparation and a big financial investment. At the same time, they have high expectations for the application and, most of all, for the result and the savings they’ll make on their translations. This attitude often leads to a reluctance to try out MT “untrained” and use it in its generic form for day-to-day business. But, as with sport, it doesn’t have to be Olympic gold right from the get go. We’ll show you that great things can also be achieved with sports shoes that have run their course.

Many still assume that an MT system is only useful if it’s customised, i.e. if it’s trained specifically for a particular company. However, using MT doesn’t always have to go hand in hand with complex data curation, training cycles and the high investment costs that come with that. Even generic MT systems – meaning machines that haven’t been specifically trained to meet the requirements of a company or specialist area – can deliver very good results, depending on the subject, text type and language direction. To draw a comparison with sport: if you want to try one out, you don’t need to buy all the professional kit straight away (even though many people do exactly that), but just first give it a go in your existing sports gear or using borrowed equipment. If you require translations for texts in standard language and standardised continuous texts, but you’d like them to be done quickly and cost effectively, it makes sense to use generic MT systems to make savings “off the cuff” and to try out the MT process for yourself.

In contrast to customised solutions, generic MT systems are available immediately and don’t require any time-consuming and resource-intensive training process. This allows companies to access foreign-language content particularly quickly, which can mean significant savings, especially with large translation volumes. However, when using generic MT in a corporate context, you should always commission post-editing. This is where a specialist human translator reviews the MT result, comparable to the training process in a new sport, where a coach provides instructions and checks what you’re doing. Because even if you were super keen to take up a sport, there’s no way you’d pick up a hockey stick or even try show jumping without help or knowing the rules. Instead of translating the entire text manually, post-editors receive the machine pre-translation to check and correct it. The MT result provides a solid basis for all the necessary adjustments to be made. Compared to human translation, it often provides a significant time and cost advantage with an end result that’s usually of comparable quality.

Compared to generic MT systems, company-specific trained MT solutions enable the machine to be provided with company-specific rules and requirements as well as specific technical terms. This increases the quality of the MT output from the outset, reducing the manual work of post-processing. However, as with the Olympics, getting there isn’t always easy and definitely requires resources and time that not every company wants to spend. Fortunately, there are a few intermediate stages between playing a sport recreationally and becoming the world’s best, which improve your result step by step and get you closer to the winner’s podium.

The first of these is to assess the feasibility. Just because, technically, virtually any text can be machine translated, it doesn’t always make sense in terms of quality and the amount of rework involved. Factors such as the MT system, the language combination, the subject and the quality of the source text play a major role here. In principle, no text can or should be categorically ruled out for machine translation: for example, a marketing text into English can deliver very good results, whereas extensive post-editing would be necessary for French. When it comes to software, context-free GUI texts will probably lead to many errors in the pre-translation, while longer error messages provide good results. It’s therefore important to consider this on an individual basis as part of a feasibility analysis in order to check where using MT offers added value.

The quality of the source texts can be optimised in advance by using translation-oriented writing, which also benefits the machine translation output. Because if the source text has clear, precise sentence structure and no unnecessarily complex formulations, there will be fewer machine translation errors. This optimisation also includes standardising the formatting and sentence structure to avoid incorrect segmentation. In this way, these measures can improve the translation quality without having to make any adjustments to the MT solution. This allows companies to increase the efficiency of machine translation and minimise misunderstandings with little effort.

Creating and integrating a glossary is an extremely effective method of optimising the quality of generic machine translation. A glossary contains the most important technical terms and their foreign-language equivalents, i.e. the specified translations. This input enables machines to translate technical terms according to the company’s specifications instead of using the most likely translation based on the training material. Integrating specifications can be pivotal for the quality of the pre-translation, especially in areas with very specialised and industry-specific terminology or for companies whose terminology deviates greatly from the standard.

And, as mentioned above, in the end, there’s no way around machine translation and post-editing (MTPE) in a professional context if you want to combine the best of both worlds: the speed and efficiency of MT and the precision and expertise of human post-editors. Professionally reviewing the machine pre-translation ensures that all errors and inaccuracies in the machine translation are corrected efficiently, while nuances, technical terms and stylistic subtleties can be fine-tuned. However, post-editing isn’t only crucial for generic MT output: just as in sport, it’s important to have someone – in sport the coaches, in MT the post-editors – checking and correcting everything to ensure you’re on course for your best performance. Post-editors can categorise and evaluate the errors that company-specific MT systems make, helping to further optimise the system with targeted feedback. Overall, however, the effort required for post-editing should be significantly lower for trained machines than for generic MT.

So, machine translation is similar to sport: you don’t have to set off to the Olympics straight away! Companies don’t need to use the most expensive solution from the outset in order to try out machine translation. Even generic MT can offer great added value, speed up processes and reduce costs. Companies that put off starting to use it for a long time because they want to be able to rely on curated data, perfect glossary specifications and optimally trained machines may miss out on competitive and cost advantages. Especially in times when speed and efficiency are crucial for serving as many markets and requirements as possible. Machine translation definitely makes it possible to translate large volumes of text in the shortest possible time and is therefore also an interesting option for areas in which frequent updates or multilingual content is required, such as in e-commerce, software localisation or technical documentation. Even if generic MT is far from perfect, the quality can be meaningfully improved by implementing additional measures, such as post-editing or integrating company-specific glossaries. Whether you’re after championship gold or a certificate of participation in a local tournament, the key is to take the first step and make the most of the opportunities already available to you.

Would you also like to get started with machine translation or do you have any questions on the subject? Then get in touch with us now.