Machine translations: technological miracle or just hype?

The market for machine translations is more varied than ever. But do these systems also deliver the results necessary to relay readers textual content from another language in an accurate and understandable manner?

There is certainly no lack of enticing promises here. In April of this year, for instance, various chambers of industry and commerce announced that the EU Commission’s online translation tool could now be used by small and medium-sized enterprises for free. Confidentiality was highlighted as a big advantage, since this tool does not collect any data for further use. There was little talk, though, of the disadvantages, which we would like to explain a bit here.

There are also many other free and commercial tools, some of which are already very well-known, while others are more obscure.

Tools that deliver machine translation can rapidly transfer large amounts of text into another language practically at the click of a button. Now and then, these tools are sufficient for gaining a rough overview of the content of a text. One should exercise caution here, however, since the resulting content may contain gross errors. This is why the EU Commission even warns on its own site that its translation tool only provides a rough translation that will require editing by a qualified translator if a precise, high-quality text is needed. Importantly, this applies not only to the translation tool from the EU Commission, but also to all other tools.

Early machine translation systems that used rule-based or statistical methods generated results that could often be described as amusing. But a real breakthrough arrived around six years ago in the form of neuronal machine translation (NMT). NMT is a subfield of artificial intelligence based on neuronal networks. The translations produced with this technology read far more naturally than the output of earlier systems. The software used for this, referred to as the engine, is fed data from existing translations, enabling continual optimisation.

The inner workings of the engine are something of a black box, however. The process behind the translation results is barely comprehensible, making it impossible to control. This drawback is naturally reflected in the results – once the initial amazement at the well composed sentences has subsided, a closer inspection reveals that not all that glitters is gold. NMT has its pitfalls. While the mistakes from earlier systems were quite obvious, it is very easy to overlook NMT errors because the sentences often sound quite good even though they are patently wrong content-wise.

Customers looking to efficiently deploy machine translation need the right technical evaluation of its limits and potential in accordance with the requirements and purpose of the envisaged communication.

Depending on the text type, machine translation presents other difficulties, and its advantages decrease as the complexity of the requirements increases. It can generally handle short, simple sentences in everyday language well. The careful use of specialist terminology, on the other hand, will often fail due the aforementioned unpredictability of the translation results. For this very reason, the system cannot guarantee uniform language use, which can leading to instances of misunderstanding for the reader. Other kinds of difficulties arise when it comes to texts meant to entertain with humour or metaphors. Such stylistic devices throw the machine for a loop, and deliberate ambiguity or allusions can leave it completely “flustered”. No algorithm to date can read between the lines. Both texts that are appropriately composed to entertain the audience and SEO-optimised texts are thus not suitable for the use of machine translation.

Machine translation can offer valuable support for certain standardised text types. This can potentially provide time and cost savings to you as a customer. However, every machine translation must be checked by a qualified professional.

We offer editing of machine translation results, referred to as post-editing, as an extension of our portfolio. With post-editing, we ensure that machine-translated texts end up speaking your language. This requires an initial consultation to determine your exact requirements for the target text. We can additionally advise you on whether the use of machine translation is appropriate or economically feasible for your texts.

If you would like to delve deeper into this topic, our articles “Post-editing in accordance with ISO 18587” and “Competences and qualifications of post-editors” are sure to provide you with interesting insights.