>>147829699

the implication would be that the act of subbing would be no longer be necessary. if the script were provided a computer would translate it automatically, probably with some editing afterwords.
the big picture allowing you to have a separate program that translates the audio live. along with conversations with any foreign language destroying language barriers.

Isn't that what neural networks are about?
It shouldn't be hard to outdo "professional" translators for anime.

This is pretty important news for people on Cred Forums who can't read Japanese. I mean I use translation aggregator and agth for visual novels so whenever Google improves it's translations it at least affects that. Hopefully they get to the point where they can do this on the fly with speech.

>10 years from now we will have tons of EOPs reading crappy machine translations and claiming to understand the story

Neural networks should be able to get there over time even with strong AI. It will just require alot of training data and possibly specific neural networks for certain languages to certain languages. Context shouldn't be a barrier for a neural network either. Through training data it should be able tlo learn when context should be preserved.

How is that different from now?

As opposed to EOPs reading crappy scanlations and claiming to understand the story like we have now?

>EOPs treating this as anything impressive
Kek.

We already have that shit today, plenty of people are complete morons like this guy, who spoil things for themselves.

The time period you're talking about is a fantasy. It'd be far easier to construct an AI, teach it all languages, make it consume the sum of humanities knowledge so it understands references, and then have it translate on the fly, than it would be to code in every single set of parameters on a neural network.

Can't we just teach the AI to program itself?

>The time period you're talking about is a fantasy. It'd be far easier to construct an AI, teach it all languages, make it consume the sum of humanities knowledge so it understands references, and then have it translate on the fly, than it would be to code in every single set of parameters on a neural network.

at least look up what neural networks are before you start coming up with bullshit

It's the difference between throwing shit at a wall to see what sticks, or waiting for the wall to grow a mouth so you can tell it what you want.

NNs will be useless until we reach the breaking point of information flow that allows sentience to develop.

What these retards are doing right now is akin to trying to get snails to understand human speech. Wait until you've got an actual base before you try and work on the big problems.

Neural networks defeated that Korean at Go.

A snail could beat a gook, they're sub-human.

The processing power required to place a few stones on a board is orders of magnitude lower than understanding speech.

7 billion people on the planet can understand speech but there was only one who could have taken a game off of alpha-go.

Speech is harder than go, but the gook that was beaten was the best at the world at a game that could not be brute-forced

Try looking into deep learning, dumbfuck.

And sentient computers are an impossibility using currently existing paradigms. NN's learning, and even deep learning, uses structured (human assisted) learning to "give sense" to things. No computing paradigm is yet able to infer and manipulate abstract concepts on his own.

>takes EOP seriously
>Kek
Jesus.

You'd be surprised. Translating literature between Russian and Czech for example can be an ordeal because Czech has an inherent sarcastic and laconic tone that Russian doesn't.

You don't need to understand all references to translate things. You honestly don't expect for every translation to have translator notes explaining every joke/reference to you, do you?

>Wait until you've got an actual base before you try and work on the big problems.
The current "base" right now is phrase-based machine translation, which was literally the first sentence of the article. Though that "base" can only translate words and poorly attempt to arrange them into sentences. That's why they're trying neural networks now.

When it comes to language, even at it's most complex, it boils down to pattern recognition and with the occasional stupid exception here and there. When it came to Go, it had over 2.0 x 10^170 legal positions making it pretty much infeasible to brute-force.

that's basically what neural networks are. you give the program a very large set of questions along with the correct or best answer and it re arranges itself until it can come up with the correct answer to a new question on it's own.

youtube.com/watch?v=qv6UVOQ0F44

I can't believe people are falling for this bait

>code in every single set of parameters on a neural network
lol you've never worked with machine learning tech have you?

Here's two critical pieces of information:

1. Google Translate was not the first or best machine translation service around. There were/are analytical engines made with a lot more linguistic expertise. Google Translate uses an inferior statistical approach: there are texts in language X, there are same (or similar) texts in language Y, let the machine decide which words and phrases match to maximize the total number of matches across the whole set. Obviously, some word form and part of speech parsing also happens, but the main principle is that simple, and anyone who has learned another language should be able to give a couple of examples that can't be translated like that, even with additional external data about the context (website categories, etc.) Google has only been able to get away with such primitive offering because of its enormous processing capabilities and inertia, killing most of the market for better options while they were at it.

2. “Neural networks” is Java-Enterprise-Solutions-Silver-Bullet of modern time. In latter case, providing complex ad-hoc mess of one-size-fits-all components that, eventually, no one can fully understand or control was found to be faster and easier than analyzing the task and making a program for that task. Neural network is even more complex (and performance-hungry) silver bullet at which you throw inputs until it magically starts producing a certain ratio of valid results. The difference is that no one knows how it actually works from the very beginning. And the talk about cybernetic systems producing coherent translations is older than computers themselves, read something from the 50s, and compare those promises to that one.

P.S. (or maybe the main point)
In fact, who uses Google Translate? It is unsuitable for any professional need. Even us addicts of the masturbatory parchments from the Orient need a better translation to get the story. Those who use it either don't need anything better or understand it exists.

I-I'm still fine aiming to be an interpreter/translator right?

I just read an article showing that their algorithm have an english bias. Meaning a French-> Italian translation goes through English as an intermediary, resulting in fuckloads of errors.

>no Maki

Depends on what languages you specialize in. Conjugation and inflection heavy languages are a nightmare for machine translation algorithms, so if you work with those you're safe.

Where does techloligy fit into that picture?