Monday, December 16, 2019

Effective Post-editing Machine Translation Techniques by Ashley Halsey

Technological developments are having a bigger and bigger influence on languages. Translation, transcription and language learning are all revolutionized by machine learning algorithms. As a result, linguists must themselves evolve and adapt their skills at breakneck speed to these developments.

Machine translation (MT) is a fine example of this kind of development within linguistics. MT was developed to translate huge amounts of text from one language into another. Although MT isn't new, in recent year it's taken huge leaps and is increasingly dependent on artificial intelligence, big data mining and cloud computing. Where Statistical Machine Translation – a translation process based on the grouping of certain core words – has traditionally been used since the mid-2000s, a new system called Neural Machine Translation has now become more prevalent, and the degree of accuracy in translation services has become greater still.

Once the translation is complete, linguists are still faced with the need to edit the output. Since this cannot be performed effectively by machine, it can be an incredibly time-consuming activity, as well as just another example of how linguists have had to add a new skill to their arsenal. But how do you become a master post-editor?


Make Decisions Quickly


One of the core competencies in becoming a highly effective machine translation post-editor is quick decision making in terms of the quality of what you have in front of you. Does it require only minor rejigging, or will it be better deleted and started again from scratch? Indecision here simply delays the process and often compromises the quality of what remains. 

“Some editors, and those who commission them, employ a three-second rule here: three seconds to ascertain if any mistakes exist, and if not, then a quick move along to the next line is recommended. This kind of speed editing does not always guarantee completely error-free text, but shows the most faith in the machine translation, thus making an investment in this technology most worthwhile,” says Audrey Kavlos, a translator at Writinity and LastMinuteWriting.

Decide on Voice

Even the most stagnant of text has some type of discernible voice which can be detected from the combination of words that are used, as well as punctuation. MT can leave a text almost devoid of any recognizable human voice, but this in itself is not an issue. What is relevant is how the post-editor maintains consistency in the text that remains (this is as true with concepts and terms as much as with voice). Any heavy tweaks will lead to inconsistencies in voice, so some semblance of continuity will need to be employed. This is not an all-or-nothing approach but is rather another example of a skill which exists in the art of post-editing.

Don’t Over-edit

Over-editing is a minefield. Once you begin over-editing, you have entered the realms of an activity which is possibly more time consuming that re-writing the entire text from scratch in the first place. It messes with the aforementioned voice and is a deeply frustrating activity that ends up compromising the original text in some way, which brings us on to…

Stay True to the Original Source Copy

Any linguist worth their salt appreciates the importance of staying true to the tone and character of the source text. It is not the job of the translator to adapt such things but instead give a faithful representation, in another tongue, of what the speaker (or writer) of those original words was trying to convey. 

“Even small changes applied inaccurately can have a profound effect upon the translated text in terms of how faithful it is the original source copy, and a delicate touch is the mark of a talented linguist," says Matthew Holderness, a linguist at DraftBeyond and ResearchPapersUK.

Don't Omit or Add Anything

Similarly, nothing can be omitted or added which was included (or not) in the original text. Again, it is not the job of the translator to decide upon the value of certain concepts, and to include them (or not) on a whim, but to instead remain faithful to that original text. Any glaring or subtle omissions as a result of the machine translation must always be rectified.

Ashley Halsey is a  professional writer, marketing expert and tech enthusiast. You can find her contributing her insights and expertise at LuckyAssignments and GumEssays.