The practice of post-editing machine translations by humans has started to become more and more prevalent in the translation industry. At first, post-editing may sound like a cost-effective solution that produces fast results; however, it carries several hidden risks for customers and service providers alike. In this article, we wish to draw attention to these dangers.
Machine translation quality
With the rise in the volume of texts to be translated, the need for improving translation speed and efficiency has also increased. One approach to this is utilising machine translation, which used to be frowned upon almost universally due to its poor quality but is starting to spread even among professionals—including translation agencies—, so much so that multiple well-known computer-assisted translation software programmes support the use of machine translation plugins. (Koponen, 2016)
Recently, machine-translated texts have seen significant improvements in quality thanks to the use of state-of-the-art artificial neural networks, and the fact that the outputs sound more natural suggests that this will most likely become the dominant strategy in the machine translation industry (Oliver, 2020). However, differences in quality exist even between neural machine translation services, as demonstrated in the following example:
Original sentence: Az utószerkesztésen átesett gépi fordítások sem ütik meg az emberi fordítások mércéjét.
First machine translation: Post-editing machine translations also do not meet the standards of human translations.
The simplicity of the style makes the translation sound unnatural; moreover, the present instead of the past participle form of the verb in “post-editing machine translations” misinterprets the subject as the one doing the post-editing instead of the one having undergone it, and using “also” in a negative sentence is grammatically incorrect—all of which constitute severe mistakes.
Second machine translation: Machine translations that have undergone post-editing are not up to the standard of human translations.
The translation’s easy-flowing and interpretative style facilitates a much more natural read; however, as a result of “have undergone post-editing” mirroring the structure of the source text instead of the much more elegant “have been post-edited” and owing to the complete omission of “sem”, it is at best on a par with a translator who is more or less good at phrasing sentences but pays little attention to detail.
One of the further disadvantages of machine translation is that even engines using neural networks cannot take context into consideration (Oliver, 2020), which results in machines not recognising cases when homonymous, repeated sentences have to be put into the target language differently. For example, depending on intention, “Fejezze be!” can be translated as “Stop it!” and “Finish it!” as well.
The dangers of post-editing
The above example shows that even with the development of technology, machine-translated texts are still not good enough to be published to a wide audience (Koponen, 2016), mostly due to the lack of lexical variety, the structures closely reflecting how the original text is composed, and the disregard for the big picture (Oliver, 2020). Anybody who has ever used any kind of automation knows that machines on their own cannot produce a quality that resembles what humans would create: AutoCorrect makes mistakes when correcting text messages, character recognition software programmes misrecognise words and clutter documents with editing errors, and speech recognition applications process human speech into written format inaccurately.
Machine translation is no exception to such mistakes, either; therefore, human intervention, that is, linguistic experts post-editing machine-translated texts, is indispensable. However, as machine translation still has not reached the level of conveying the meaning of source-language texts truthfully (Koponen, 2016), correcting the grammar and style of the target-language version by means of monolingual post-editing is not enough, the content of each translated sentence also has to be checked against that of the original. This means that there is little difference between proofreading the work of translators who produce poor quality and post-editing machine translation.
Post-editing may seem like a cost-effective method because, as opposed to human translation, which strives to achieve an easy-flowing and natural sound, its sole purpose is to create presentable texts from pre-generated content. Nevertheless, it should also be taken into consideration that the machine translation of certain sentences tends to be so inaccurate that it would be easier to translate them from scratch (Teixeira, 2015) than to try to adjust the automatic output to the source-language text, which is illustrated well by the machine and human translation of the following sentence:
Original sentence: Rendezte sorait, mielőtt nyugdíjba vonult volna.
Machine translation: He arranged his lines before he retired.
The machine produced a grammatically correct but nonsensical, word-for-word translation. The reason for this is probably the fact that its database does not contain enough idioms for it to recognise the non-literal meaning of sentences easily.
Human translation: He had his ducks in a row before his retirement.
The human translator, who is familiar with the figures of speech used in Hungarian, was able to produce a translation that fits the culture of the target audience well despite the fact that the idiom in the original sentence does not necessarily have an identical target-language counterpart.
This means that post-editors practically do the same job that translators who do not employ this kind of automation do, just at a lower price. Moreover, during the process of post-editing, personality, which sets good translators and even better translators apart, is lost because linguistic experts have to rely on the outputs of machine translation (Oliver, 2020) combined from different sources, while making sure that the style of the sentences that need major revisions fits the rest of the text.
In order for post-editing to produce a result that is acceptable in terms of quality with as little effort and as quickly as possible, it is important for the original text to be available in a format that the machine can process (Tirosh, 2018) and for enough reference texts to exist in the given language pair and field that the machine can rely on. This is usually achievable between major languages and in the case of—mostly legal and technical—texts that contain frequently recurring phrases. In the case of minor languages and texts that have fewer fixed expressions or that have been written by non-native speakers, creating a content suitable for human consumption would often require so much work that it would be more cost-effective to leave the entire translation to an experienced professional translator and not use machines.
Instead of post-editing, choose our trustworthy professional translation services—you will get more value for your money, and as a result, you will receive a top-class text that is tailored to your needs in all respects.
References
Koponen, M. (2016). Is machine translation post-editing worth the effort? A survey of research into post-editing and effort. The Journal of Specialised Translation, (25), 131–148.
Oliver, A. (2020). Human translation and machine translation. Specificities, uses, advantages and disadvantages. Linguapax Review 8. 111–129.
Teixeira, A. (2015. 03. 31.). I Don’t Offer Machine Translation Post-Editing Services, Here Is Why. Retrieved 29.09.2019, source: Anthony Teixeira – Professional French Translator: https://www.at-it-translator.com/i-dont-offer-machine-translation-post-editing-here-is-why/
Tirosh, O. (2018. 08. 30.). Post-editing machine translation – all you need to know. Retrieved 24.09.2021, source: Tomedes: https://www.tomedes.com/business-center/post-editing-machine-translation-need-to-know