Machine Translation

An interesting thought I’ve always had was why translating from one language to another a big deal? Why would there be a need for translators if you can just have a word bank? My first thought had been that since dictionaries exist, you could hard code the translator to just replace the word in one language with another word in another language, according to your dictionary, and problem solved you would be able to translate the sentence. The first problem with that naïve thought is that languages are different (crazy, right?) in more ways than just spelling or definition alone. The way they handle and communicate subjects and objects, can be vastly different, and the way they treat sentence structure can also be different to the point where the end result does not have the same structure as the original statement. But, ultimately, the goal should be so that the receiver of the message gets the intended effect from the speaker, not necessarily that the original message has the most similar meaning, because these are two subtly different things.

The reason this goal is important is because even if you are close when translating a language, there are implications and ambiguity of the meaning that may be difficult to capture. The best example I can think of is the phrase, “Thanks a lot”, often used to stress an annoyance or hindrance someone has caused someone else. This is often used as sarcasm, but what makes this hard to translate is that it’s often used as a riposte, and making a curt and to-the-point comment might not necessarily be “thank you very much but not really” in another language. This fact and the act of figuring out context to determine what the intended definition is makes machine translation (MT) difficult, since to definitely pin-point what the correct translation method is it would have to deal with both of these problems.

Something to think about is that language is not closed to only language, and meaning takes influence from the surroundings, like visual cues and sound. It wasn’t apparent to me when I first though of translation, but it is clear when we think about our interactions with the people around us. Our tone changes, we make hand gestures, we have different facial expressions, etc. These cues, while maybe not easy to translate, provide guidance for what a speaker intends to communicate, as well as what is implied when used in conjunction with language. So, while it may never be possible to completely capture the meaning from a speaker, just translating text would be challenging since MT would need to both figure out context so that it could actually communicate what a speaker intends, and be able to use that context to discern the meaning of each word in each sentence to get the same result in the target language. They may not be as good as human translators, but the technology is improving, and just like my first naïve idea the technology is evolving to be the best that it can be.

Maybe one day, it will be able to fully encapsulate and translate ideas, but for now, that remains a dream for the future.

Thanks for reading!