What is machine translation (MT)?
Google Translate, Deepl, Microsoft Translator, ChapGPT... Artificial intelligence is here and as a result, the translation industry is undergoing changes like every other. Machine translation is not new to the translation industry but recent improvements in the accuracy of machine-translated texts are having an impact on the industry like never before. Neural machine translation, which uses artificial neural networks to understand and generate more accurate and fluent translations by learning from vast amounts of bilingual text data is here, and the role of the translator in the overall process is changing.
What is machine-translation post-editing?
MTPE (or PEMT) is the process of reviewing and correcting translations produced by machine translation systems. Instead of translating a text "from scratch", a post-editor reads the machine-translated text and edits it.
The main advantage of machine-translation post-editing is increased speed and thus, increased cost-effectiveness. By providing a preliminary translation that human editors can refine, the overall time needed for the whole translation process is theorectically reduced compared to starting from scratch. However, in practice, this is not always the case.

What are the potential issues with machine-translation post-editing?
Variable source or target quality: if the source text (original text to be translated) is of poor quality (garbage in, garbage out) or the initial machine translation is of poor quality (due to poor source data quality or relevance or poor training models), the post-editing work increases.
Contextual errors: machines are not as effective at taking into account context, leading to errors that may require substantial correction during the post-editing process.
Inconsistency: a key aspect of a quality translation, machine translations do not consistently produce the same translation for one particular term throughout a text.
Loss of nuance: idiomatic expressions or nuances in the original text can be missed or poorly translated by the machine.
Cultural sensitivity: cultural differences or context-specific cultural meanings are missed by the machine.
Potential for human error may increase: editing and translating are different skills; errrors made by machines are different to those made by humans and adding a further "stage" to the overal process can sometimes increase the chance of errors being introduced. Post-editors may also "over-rely on" or "trust" the MT output too much and inadvertently miss errors or corrections.
Loss of the cohesive “sound” of the text: because you are leaving some things and editing others, the writer’s “voice” does not come across and it can end up sounding less natural as a whole and not being as easy to read.
Acronyms and initialisms: these are handled poorly by machine translation, particularly if they are only two letters. Although the post-editor should pick these up, it can be very easy to miss this kind of error.
Is machine translation post-editing the best choice?
YES
For budget
MTPE can be more budget-friendly compared to full human translation, as it leverages machine translation for the bulk of the work. However, the cost-effectiveness depends on the quality of the machine translation and the extent of post-editing required, as well as the skill and experience level of the post-editor.
For speed
If speed is crucial, MTPE can be advantageous but this comes with the risk of a loss of quality.
NO
For complexity
For content where high precision and nuance are critical—such as medical texts, legal documents, or marketing materials—full human translation can often be preferable. MTPE can work, but it requires highly skilled editors to ensure high quality. The time taken to post-edit this type of text may not be any faster (and could actually take longer) than a highly-skilled human translator translating from scratch.
For hand-written text
Machine translation cannot translate hand-written text.
For cultural understanding
Machines are not effective at incorporating cultural understanding into translation of texts and will probably cause far more problems than they are worth.
Next up... an in-depth look at some of the issues we introduced here, and how to address or mitigate them. See you in the next one!
Comentarios