by Reinhild Möller
Machine translations may be getting better and better, but they are still far from being flawless with no human touch. The standard ISO 18587 sets out the requirements for post-editing of machine-translated texts, but it explicitly covers only full post-editing. In practice, though, the process is influenced by the competing factors of high time and cost pressure on the one hand and high quality standards on the other. To manage this balancing act, translation service providers and their clients can negotiate standards for light post-editing.
The past few years have brought ground-breaking advances in the field of machine translation (MT). The technology of neural machine translation (NMT) in particular has elevated the quality and readability of automated translations to a whole new level. This has made high-quality MT systems a helpful tool that can be used to complete certain translation projects on a short timeline and with a low budget.
These systems are not free of errors, however, and the MT results must be post-edited. This is when professional human translators ensure that machine translations fulfil customer requirements regarding both specialist terminology and style.
Machine translation is growing in importance as the quality of the results steadily improves. This led to the publication of a standard in 2017 that specifically details the requirements for post-editing of machine translations: ISO 18587. It is meant to be used by translation service providers, their clients and post-editors.
ISO 18587 defines post-editing as the editing and correction of machine translation output. The human translator who edits the MT results is referred to as a post-editor. The standard fundamentally differentiates between full post-editing and light post-editing. However, because the standard cannot cover the full range of possible quality levels for post-edited machine translations, it only lists the requirements for full post-editing. According to ISO 18587, MT results should be “comparable to a product obtained by human translation” once they have undergone full post-editing.
As with the translation process outlined in the ISO 17100 standard on translation service requirements, the post-editing process is divided into pre-production processes, production processes and post-production processes.
The pre-production processes begin with the translation service provider determining – along with the client – whether the relevant source text is appropriate for machine translation. The quality of the MT results depends primarily on the quality of the MT system as well as the language combination, the text type, the subject area and the style of the source text. The effort required for post-editing should be less than the effort required for a new translation.
If the MT results are usable, then the translation service provider and the client must determine the post-editing project requirements (quality level, target group etc.). Post-editors must receive access to all reference materials and any resources required for post-editing. Since MT systems are ideally used in combination with translation memory (TM) systems, post-editors must be able to clearly distinguish between MT and TM output.
During the production processes, post-editors must ensure that the post-edited MT output is comprehensible, that the source language content corresponds to the target language content, and that the project requirements are fulfilled.
The standards for fully post-edited MT output largely correspond to the standards for a human translation:
With regard to the semantic (content-related) correctness, however, post-editing sets special requirements that differ from those of a human translation. This is because there is already a suggested translation to be evaluated. According to ISO 18587, post-editors are responsible for:
Unlike translation projects that do not involve MT, post-editing requires post-editors to juggle three different texts: the source text, the MT output and the final target text. Post-editors are supposed to use as much of the MT output as possible on the one hand while fulfilling the standards of a human translation on the other. Post-editors must therefore:
During the post-production processes, the translation service provider verifies that all specifications have been met and releases the post-edited output for delivery to the client. Post-editors should also provide feedback on the performance of the MT system so that it can be improved further.
ISO 18587 lists possible elements of project specifications and agreements between the client and translation service provider in an informative annex. For example, the scope of the project should be agreed in advance: full or light post-editing, check by the post-editor (single control principle), revision by a second translator (dual control principle), specialist review, transcreation etc.
The ISO 18587 standard specifies the process and requirements for the post-editing of machine translations, thus creating a common basis for all participants in this process. The definition of important specialist terms ensures that everyone involved is speaking the same language and has a common understanding of these terms. The clear definition of full post-editing ensures transparency and security. All participants have the same expectations of the fully post-edited MT output – it should be comparable with the work of a human translator. The quality of the post-editing results can therefore be evaluated by the same criteria used for regular translations.
If you would like to learn more about the professionalism of post-editors, read our blog article “Competences and qualifications of post-editors”.
Receive useful information relating to translation and interesting project reports straight to your inbox.