In the past I used to be worried about machine translations and what they could mean, but ultimately the limitations of the programs and algorithms used right now can in my mind only be overcome with an ability to comprehend and grasp the context in which a text finds itself in. Literal translations for one, are not always the best translation, very rarely so if I had to say so myself. And if there was an AI that could understand and process enough context to produce a translation like a human would, that is, considering factors like how the response the words it chooses to use would have on a potential reader, then in my mind there are way bigger problems to think about than human translators being obsolete. Humans in general would be obsolete if it ever came to that.
A good example would be something simple, like to be or not to be: that is the question, and from GoogleTL and DeepL are the following translations:
生きるべきか、死ぬべきか：それが問題です
To be or not to be: that is the question
Machine translation being what it is now, mostly drawing from already existing translations, popular searches like Shakespeare default to mostly what is the most popular translations, and you can clearly see where it adds its own flair with the desu ending, which is a very odd choice for the context, while DeepL just outputs English. the following line, whether tis' nobler in mind to suffer the slings and arrows of outrageous fortune, Or to take arms against a sea of troubles, And by opposing end them? results in the following, from GoogleTL and DeepL, respectively:
苦しむ心の中で気高いかどうか。 とんでもない幸運のスリングと矢、またはトラブルの海に対抗するために、そして反対することによってそれらを終わらせるために？
悩むことが心の中では尊いのか？悩む方が心の中では崇高なのでしょうか？それとも悩みの海に武器を持って立ち向かい、それを終わらせる方がいいのでしょうか？
Which, in my mind, displays the limitations of machine translation in it's current iteration. The GoogleTL translation is 95% of an already existing translation, with particles and such changed, while DeepL defaults to what it does best, word-for-word translations of nouns and verbs and recasting into English word order, except it makes a mistake and outputs the first line twice. More than anything this is likely due to the fact that GoogleTL has had more requests for that particular line than DeepL does, causing it to default to the most popular translation plus or minus a few particles, while DeepL gets none and therefore defaults to a 'proper' machine translation.
The faults of GoogleTL are that it currently, and DeepL also to an extent, work on existing translations without full context, and without existing translation it defaults to translations that have the same problems (albeit less of them) that machine translations have always had.
Steven C.
No comments:
Post a Comment