BITS and Pieces

Get global. Get ahead.

Machine Translation: Behind the scenes

As the name suggests, machine translation implies algorithm-based translation of a text into another language, done completely by a machine. Now on the face of it, it sounds like a wonderful concept. Who wouldn’t like it if a machine could directly translate everything for us like in Sci-Fi movies and stories? But like the other technologies of this genre, a perfect machine translation is also nowhere to be found in the near future. Right now, what we have is a process called “MT–Post editing” which means machine translation and post editing. Google translate is a viable option for personal use, or for understanding the bare bones of a text, but anyone who has used it knows that in matters of importance, it is better not to rely on it completely. So here is a rather brief and simplified explanation about MT-Post editing.

What is MT-Post Editing?

Simply put, this process entails a document being translated by a machine (dedicated software and programs) and then a human reviewer goes through the entire file to verify this translation. In particular, the editor has to keep in mind the limitations of machine translation. For example, it is likely to base its translations on glossaries and other sources of bilingual data, which prevents it from taking the actual context of a text into consideration. It only sees a sentence as a sentence, and this might cause problems related to readability and Word Sense Disambiguation (one word can have several meanings). Moreover, idioms, language-specific characteristics, and colloquial terms are more than likely to trip up the machine translation program. Therefore, the human editing is as important, if not more, than the actual machine translation.

How does it work?

The most used method for MT today is algorithm-based (google has just launched its neural translation campaign, but it is not relevant for commercial translations, yet). So in layman’s terms, companies create and sell huge glossaries and databases containing bi-lingual (or even multi-lingual) terms, phrases and sentences. The MT programs use these databases to cross-reference meanings and give the end user the ‘matching’ translations. In this way, it can work significantly faster than a human can, but often compromising on the qualitative aspect.

Where is it used?

Based on the previous paragraph, it is easy to see which fields will benefit the most and the least from this technology. On the one hand, straightforward texts where a lot of the matter is standardised or quoted, like legal contracts, patents or even instruction manuals, will be relatively easy to translate using MT. On the other hand MT will be more or less a hindrance if used to translate literary texts, where one has to read between the lines more often than not; the same holds true for website content, correspondence, advertisements and so on.

Is it going to replace human translations anytime soon?

Suffice it to say, there is a long way to go before humans are made redundant in the field of translation. To truly replace humans, it would take an artificial intelligence which is on par with ours, which can reason instead of following a set of rules, which can think logically and then apply context to a text. As of now, machine translation is rule-based and there are no fixed rules or algorithms in languages. Cultural references, satire, irony, humour, or any other such aspects of language are completely lost on machines and could result in serious consequences if passed off as “true” translations.

Should I opt for it?

The use-case for machine translations is a sort of a niche as of today. As a potential translation client, you need to have a certain type of documents for preferring MT to a human translator. Certainly, the human will take more time and in all probability, more money for the same translation, but the quality would be significantly better, with all the context of the document being understood and represented in the target. So the answer would be, for translations that do not have a very high quality requirement, which are more or less standardised and straightforward, and preferably not of a critical nature, MT-post editing is a godsend, which will improve processing speed and effectiveness dramatically. For all translations other than these, human translators are still the safest and most effective option. So there’s really no getting rid of us!

Back to the Main Page of this month’s issue >>

Sushant Bothe - French-English Translator, BITS Private Limited A curious and headstrong mix of patience and aggression, creativity and discipline, and caution and recklessness, Sushant is a mixed bag of tricks if there ever was one! A myriad of passions govern his life, like football, bikes, languages, technology, psychology and good literature. He’s also a gadget junkie and an amateur marksman. His ‘Kryptonite’ is monotony and getting up early in the mornings! An introvert, he usually minds his own business, till someone gets a fact wrong or makes a grammatical error.