We live in a time of ‘posts’: post-modern, post-human, post-truth, post-capitalism. How about post-editing?
Translations generated by a machine and amended by human beings: post-editing is a term that has been on the lips of many in the translation and localisation industry for a decade or more. In its infancy, post-editing of machine translations (or PEMT, as it is also known) was derided as a strategy for cowboy translation firms to churn out low-grade content on the cheap. With the technology having since caught up to the quality standards expected in the industry, PEMT has, for the past few years, been quietly revolutionising the way that agencies, freelancers and businesses tackle everything from legal documents to financial reports. This is not simply a new trend or add-on; PEMT is an all-new professional field, with an increasing amount of accreditation options available to those looking to specialise in it – there is, in fact, an ISO standard for post-editing of machine translation output (ISO 18587:2017).
But what is post-editing, exactly? And why is it so revolutionary?
Put simply, PEMT allows translators to deliver work quicker. Instead of translating a text from scratch, a machine tool provides a translation ‘blueprint’, which the translator then stylistically and grammatically nips and tucks until it comes close to a human translation. A word here, a phrase there, a restructuring to bring the syntax more in line with the target language: it is a genuine task and one that calls for eagle-eyed observation and an understanding of the nuances of the source and target languages. Some texts are more suited to the process than others: legal documents, operating manuals, corporate communications and technical texts are among those that most regularly undergo the PEMT treatment.
The figures speak for themselves: according to a joint study by the University of Groningen and Dublin City University, PEMT improved translation productivity by 18% compared to translating a text from scratch, while neural machine translation – which uses AI to help shape the text – increased efficiency by as much as 36%. On top of that, the translators who were tasked with post-editing the text took between 25-50% fewer breaks than those who had to create a new text in its entirety.
For companies, this decreases the text’s time-to-market significantly while reducing their overall translation costs. This is technological advancement with the human touch. Everyone stands to benefit – just as long as they follow a few ground rules.
Efficiency, but not at the expense of security
This efficiency is only worth aiming for when the end product offers a usable level of quality. Like everything in life, not all machine translation tools are created equal, and without an accredited expert overseeing the translation cycle from beginning to end, the post-edited translation can end up being just as poor as it was in the technology’s early days.
Some businesses may be tempted to bypass the involvement of professional translators and simply take care of their PEMT activities in house. After all, there are plenty of free machine translation tools online – DeepL and Google Translate being the two main players. However, the drawbacks outweigh the benefits: general all-purpose online solutions do not have the ability to learn in a targeted way. The incoming streams of information are lumped together, so the likelihood of the tool opting for the wrong vocabulary is higher than if a specialised in-house tool was used. As an article in Forbes states, a free machine translation tool “doesn’t understand industry-specific jargon […] and the wider context of a piece of content is lost on it”.
Then there is the fact that any information entered into one of these tools runs the risk of being duplicated, stolen or otherwise misused. An example: after entering the words ‘volledig ondertekende overeenkomst’ in Dutch (meaning ‘fully signed agreement’) in the free online tool MyMemory, the user can scroll down and see a host of ‘human contributions’, or extracts from other texts that have been fed through the tool. It would take only one recognisable name, date or file path for Internet sleuths, blackmailers, hackers and other cyber-attackers to trace it back to a company and potentially cause serious problems and reputational damage.
The human touch counts
PEMT is not a basic administrative task that can be farmed out to just anyone. As mentioned above, it is a bona fide skill requiring many hours of training and practice – along with a background in translation – to become adept at adapting MT texts to the required quality. In many cases, it is not as simple as changing an adjective to an adverb or adding in missing commas. For idiomatic phrases in particular, the machine tool may translate the words literally, and while they may sound syntactically correct, the meaning can be completely different to the one intended. Consider the following:
Source: Es ist fünf vor zwölf.
Machine: It is five to twelve.
Human: It is high time to get this done.
The text is correct, but the meaning is not. Similarly, the content of the sentence may be correct, but the sentence itself could sound like a machine has translated it:
Source: Der Einsatz maschineller Übersetzungssysteme schreitet voran, um den Bedarf der Übersetzungsbranche zu decken, an die immer höhere Anforderungen gestellt werden.
Machine: The use of machine translation systems is advancing to meet the needs of the translation industry, which is increasingly demanding.
Human: The use of machine translation systems is becoming increasingly widespread in response to the ever-increasing demands placed on the translation industry.
Finally, the machine tool may draw on previous translations to create a new one, resulting in additional elements that have no cause to be there:
Source: Le contrat est compréhensible
Machine: The current agreement is understandable
Human: The agreement is understandable
These three examples show how important the post-editing stage is in PEMT. Machine tools are improving all the time, but they cannot do everything – and they can become confused. Achieving the desired quality calls for a professional, no matter how good the machine translation may appear at first sight.
A time-saving secret weapon
Of course, a machine translation does not have to be pitch perfect every time: PEMT also comes into its own when a text needs to be translated for internal use only and time is of the essence. Take legal texts, for example: generally speaking, a translation of a contract will defer to the source text in terms of its legal validity, meaning it is often enough to have a workable translated text. A good example would be an estate agent in Berlin handling multiple housing contracts involving non-native speakers; rather than struggling to guide the tenant through a forest of legalese in German, they can have the contract machine translated and post-edited into English, French, Spanish, etc. If the contract does have to be watertight in the target language, the machine translation can be handed over to a legal specialist, who can then go over it with a fine-tooth comb.
The same applies to the finance industry. Virtually all banks are international in scope, with teams of employees from different countries requiring access to research reports, news and press releases in order to track market and business developments – and make informed decisions on behalf of customers and other stakeholders. With material being published by academic institutions, corporations and foundations the world over, not all of it will be available in the employee’s target language – which is where PEMT comes in. A report published by a thinktank in Switzerland on Friday could be automatically translated, manually polished and delivered into the hands of the person who needs it before the markets open on Monday – giving that business the edge over the competition.
The bottom line….and post-editing by SwissGlobal
Machine translation is here and it is changing the face of the translation industry – but it is not perfect. Post-editing is an essential part of the new translation chain; without a professional human translator casting their expert glance over the text, there can be no guarantees it will be correct or usable. What PEMT can do is handle large volumes of text in a relatively short period of time without having to break the bank for the privilege. It is certainly worth exploring and will only grow in significance as the tools – and the technologies behind them – continue to improve. Watch this space.
As a leading name in the translation industry, SwissGlobal always strives to remain at the forefront of technological developments in the field while upholding the same level of Swiss quality for which it is renowned. This is why, in addition to embracing machine translation, we have applied for ISO 18587 certification in post-editing of machine translation output. We are pleased to announce that we will officially be certified in May 2021.
Interested in discovering the opportunities provided by post-editing? Get in touch with us today and we will give you a rundown on the services we provide – from PEMT to human translation, transcreation, editing and more.
- Artificial intelligence
- Machine translation