Having translated hours of audio files as well as text of various formets (written text, transcript of speech and even the new “messenger text”,) it is without doubt that contrary to the popular hype, translation will not be fully automated in the near future.
There won’t be an app or a device that will allow fully automated real-time translation, and the following are the reasons. In order to translate a content that consist of both the apex of the cutting-edge scientific technicalities, as well as the linguistically intricate expressions that are akin to poetry, many tools were sought to aid me in its translation.
From the newest algorithmic translator utilising the big data collected in the internet, to the translation memory system which ‘learns’ how I translate a document, none have shown much promise and the following are the reasons.
1. Reliance on dictionary/word base
In most automated translation system, content is broken down into words. When an input document is given, the system may analyse the document sentense-wise (which already is at the frontier of the latest technology.) This reduces mistranslation where a single word can have multiple meanings. However at the end of the day, translation relies on a dictionary. Temporary creation of words is an excellent way to achieve efficency in high-level communication, which the Oxford-wielding bot will struggle.
While a human-being can accurately translate a sentence without knowing all the words in it, a machine cannot. Surprisingly, this is because a human can fake the knowledge of the word while a machine cannot. The critical flaw to the machine-learning translator is in fact our lack of understanding in the model which the machine was built after: human brain.
2. People don’t speak so clearly
Diction is amazing, and it is used frequently even as we speak. Medical, legal and millitary personnels use diction to document content quickly. Typing is just too slow, and audio data is difficult to access quickly. What you hear, however, is that they speak remarkably clearly, often adding in words like “full stop. comma, open parenthesis, close parenthesis.”
Diction fulfills a purpose, but not the purpose that most people actually have in mind. Any real-time translation based on verbal communication will require some form of diction, and until such system gains an ability to understand speech as human do, it will not work.
There are, of course, many devices that seem to understand our speech very well. From phone assistants to home modules, these systems are remarkably accurate in knowing what we want. Ask about wheather, restaurants near by and even current issues, and they will bring that information for you.
However, the way they make out what we’re saying is vastly different to standard voice-to-text diction. Although the diction isn’t accurate, it automatically fills the gap based on pattern recognition in order to ‘pretend’ as if it has completely understood the line. This bring us to the final obstacle.
3. We understand communication wrong
The biggest reason why current modes of machine translation does not work is because we taught them wrong. The way we have designed the machines to perform translation – and ultimately communication – was largely based on how humans do those tasks. This, is where we lack knowledge.
Algorithmically, language is made up of words and each word both in itself and in relation to others, has a meaning. It is the combination of these meanings that give rise to a content. In order to properly translate a content, instead of performing word-to-word swap or even do a sentence-by-sentence translation, one must fully assimilate the entire content and create a whole different version of it, with a whole new permutation of words in a different language.
This is an extremely difficult feat and certainly an impossible one for the computer systems in this near future.