News DeepMind Al translating is happening, human translation is becoming a history!

Sounds good and all but I'm a lazy fuck that's too lazy to use it even if it actually works properly.

 

Translating chinese to english is way too fucking difficult for any programme, trust me. If it’s a language that uses english letters (and those tend to be the easiest languages to learn) like malay then it’d be possible, but there are too many discrepancies between the chinese and english language. Way. Too. Many. Many chinese words aren’t even in the dictionary, others are a mixture of slang and stuff. And don’t even get me started on chinese literature quotes or poems. The languages are so fundamentally different that I really can’t imagine it being translated properly unless the text is simple.

I’d reckon a decade(s) at best before this becomes possible

 

I tried to use AI translation on CN web novels, and the result is not satisfying. Too many proper nouns you cannot just translate plainly. The complext syntax and casual punctuation are also problem. It's very easy to produce nonsense and logical mistakes, let alone style and taste. I think literary translation is still a zone that machine translation fails to not touch.

 

If it reaches the point where it can translate close to a native speaker, I will stop translating and instead use exactly THIS to translate my English original works to syosetu, pretend to be Japanese, and wait for my works to be adapted to manga, anime, drama CD, etc. 8D

tbh it's always better to create something on your own, so I'm not that worried if in 5-10 years from now translators aren't needed anymore.

 

Don't worry, before we get replaced, the doctors and accountants will be replaced first.

 

As someone with an accountant license... HEY!

 

To be honest some novels on LNMTL convey the original meaning better than the human translator of the series already if you ignore the fact the grammar isn't 100% correct English. Human translators can sometimes take liberties changing things to make them sound better in English but sometimes that loses part of the author's original meaning/intent in the process, whereas machine translation does it pretty much literally so while it may not sound as much like proper English it actually conveys the author's intent better.

Of course, that's assuming you're adapted to reading MTL and it doesn't look like dolphin-speak to you. I had to look up some idioms or phrases while reading occasionally to understand them properly.

And yes, I have read both HTL and MTL for parts of some novels and compared them side by side so I'm not just pulling this out of my ass.

As for whether you prefer original meaning or prefer it looking more "correct" in your language that's a matter of opinion of course, but after having gotten used to reading MTL I prefer the former personally.

 

Intent is about more than just meaning though, and there's more to meaning than the literal definition of words. The writing itself is also an important part of the story, separately from the plot/events of the story.

Reading MTL, even if the MTL is perfectly accurate (it's not there yet) and results in language that seems natural and readable (that's a long way off), is nothing like reading a novel as it's originally written. Even perfect MTL just gives you a description of the story; like reading an extremely detailed summary, or having someone describe every event in the story in their own words.

No matter how thorough and accurate it is, you're only getting information, not literature. That's why the writing is as important to a novel translation as the translation itself.

 

Well if you want to use that argument, most human translations aren't exactly art forms either.

Unless it's literally poetry or classic literature with a lot of obscure text then there's not really anything missing in MTL for me as long as terms are covered (LNMTL lets you make specific terms for each novel for things than don't get translated well, so most of the holes in the MTL can be "patched" by humans aside from minor things like grammar).

Especially for most modern CNs, there's hardly any obscure artistic value in the language itself, you're just reading it for the entertainment value of the story. And like I said for idioms and the like, I can just look them up to understand the meaning behind them. Then when they appear again in the MTL I'll understand what they mean and be able to appreciate them better than if they're translated into some English equivalent.

 

I think it will be more worthwhile for people to learn the original language (like, memorizing the specific terms that often pops up) than trying to decode the dolphin-speak. I mean, that's the best way to convey the author's intent, right?

Translation is as much of an art as drawing or any other art form you can think of, actually. Maybe some of them are doodles, maybe some of them are fine art of Da Vinci level. But when you're trying to put your personal touch into the work to convey the intent with explicit and implicit ways, that's art for me.

 

https://www.livescience.com/52207-faster-3d-computer-chip.html
https://www.livescience.com/59855-3d-computer-chip-boosts-processing-power.html
https://spectrum.ieee.org/semicondu...ps-are-destined-for-deep-learningor-obscurity

We have about 10 years left in which the semiconductor industry will scale process nodes from the current 7 to 14 nanometers to around 0.5 to 3 nm. That will yield modest improvements in computing. But it's possible that a 3D or neuromorphic architecture will be able to boost performance by up to a thousand or million times, especially for massively parallel machine learning applications. Today's CPUs are basically flat grids of transistors. If you can solve the heat issues and scale in the vertical direction using layers of transistors, you could have up to tens of thousands of cores taking up about as much space as today's single/double digit amount of cores. This is the approach being taken for the latest flash memory used in SSDs, with the current state of the art at about 96 to 128 layers. Neuromorphic describes brain-inspired architectures that use much lower power and operate in a manner similar to the human brain. The low power makes it easier to scale up vertically, because you don't have the same heat issues.

Imagine that Google's hardware improves by a million times within 20 years or so. It may be possible to completely conquer translation without the "strong AI". If not, development will continue until you essentially have a superfast human brain in a vat or on a chip, slavishly translating among other tasks.

 

I already do that when I read on LNMTL when I'm making terms or checking stuff that's mistranslated. I can probably read a few hundred Chinese characters by now just from doing that. (no idea exactly how many since I never tried recalling all of them, plus I can probably only write less than 5% of the amount I can read)

It does help a lot with understanding the original intent even if it's just from reading MTL since I often know what the original Chinese is for the terms and I usually read with the raw underneath each paragraph like this:

Spoiler: Screenshot

Still a long ways off from being able to read raw Chinese without the help of MTL though.

 

Well, if quantum computing is still beyond our reach, there always optical computing. Photonics could be an alternative, if only we could solve several design challenges associated with it...... So yeah...

 

It's not clear that quantum computing would be useful for many applications, but quantum machine learning may be possible.

Analog photonic computing may be possible. Optalysys is one of the more well known startups in that area but they are not very open about their product. They originally teased multiple exaflops level of performance in a low power desktop-sized machine, but it has since become clear that their product has narrow capabilities.

Companies like Intel are working on optical interconnects to move data between processors faster, which is probably necessary for supercomputers to continue scaling. There's also optical transistor designs being worked on, which would use photons instead of electrons on the chip.

 

Nothing is perfect and there is always flaws which can be corrected. The optimism for an expedient translation, is something growing in our age. You know what I want to see, the rebirth of dead languages.

Some may want the language barriers of current societies to become interconnected, that's cool. I wan't to see the differences of slangs and informal speeches throughout the generations.

MTL will translate most direct terms and whatever is within its database to convert.
HTL can translate based on their experience/understanding of both Original & Target languages
AITL is still in development, but for full operationality expect its use within 2-5 years.

Again neither TL is perfect, HTL was thing before 2006, MTL started around that period and in 2016, MTL gained a boost to its algorithms & library database. Following that logic AITL is currently in its infancy, it may take a while for developers/programmers to reach the publics personal standards (Business, Social, Personal, etc).

 

The day AITL can convey the meaning of

寸土为寺，寺旁言诗，诗曰︰明日送僧归古寺 and

双木成林，林下示禁，禁云︰斧斤以时入山林

is the day HTL will become useless. My suggestion is try reading less brainless stuff.