Will most people stop studying Chinese now that AI is so good?

[mouse]

On 5/29/2023 at 12:14 PM, Jan Finster said:

When....(!)

 

It is just a matter of time. I am actually surprised I come across as somewhat of an "early adopter" on this forum. Typically I am a late adopter (I typically start loving music bands or fashion when they are already considered "out") ?

 

I agree that the ability of AI to translate will continue to improve, but the idea that a real babelfish will be created that would completely replace the need to speak another language is a pipe dream. We're talking about simultaneous translation that wouldn't be hampered by wait times. I just don't buy it.

[Jan Finster]

On 5/29/2023 at 4:09 PM, mouse said:

We're talking about simultaneous translation that wouldn't be hampered by wait times. I just don't buy it.

 

I do not think speed is the issue here. Even now you can dictate something into google translator and it produces the text virtually in real time. Also, do try the "conversation mode" on google translator.

It will take a short while until accuracy is near perfect, but I would give it less than 10 years. It is from an AI perspective not an impossible task at all.

[Lu]

On 5/29/2023 at 4:30 PM, Jan Finster said:

I do not think speed is the issue here. Even now you can dictate something into google translator and it produces the text virtually in real time. Also, do try the "conversation mode" on google translator.

It will take a short while until accuracy is near perfect, but I would give it less than 10 years. It is from an AI perspective not an impossible task at all.

I agree that speed is not the issue, but quality is, and it will not improve beyond a certain point. Google Translate is now 17 years old and it still isn't all that great for Chinese. As I said in another thread on AI, I fed it a paragraph of a contemporary novel -- not a difficult paragraph, a fairly straightforward one, with correct grammar, no dialect words -- and it made several serious mistakes (and a number of smaller ones). Now take Google Translate to a corner shop with a 老板娘 from Hunan, or even Henan, and see how well it understands her.

 

I frequented the local 'language cafe' for a while, where Dutch people help immigrants with their Dutch. Google Translate is very useful for getting one's intentions across when one party speaks only Farsi or Tigrinya and the other only Dutch, English and Mandarin, but you have to be very deliberate about the sentences you make. Even something as seemingly straightforward as 'I went to the supermarket at the crossing' can become completely unintelligable when 'crossing' goes from Farsi to English to Dutch. And again, that is after seventeen years of development by a lot of smart people.

 

But that was not the question, the question is 'will people learn languages less because of AI'. But I have to get some work done, perhaps I'll be back at my next bout of procrastination.

[realmayo]

On 5/29/2023 at 1:45 PM, Moshen said:

Presumably they had warm fuzzy feelings from the exchanges also

 

I suppose I'm worried about a world in which people think you are peculiar and annoying for insisting in speaking in intermediate-level language when you could be using a high-quality real-time interpreter.

[mikelove]

I mean, why does anybody bother to learn arithmetic when you can have a calculator do it for you? (and do it absolutely flawlessly, I might add, which is unlikely to ever be the case with translation AIs) I don't think practical usage is the real driver of most language learning anyway, any more than people take up violin or piano because they want to become professional musicians; people do it because it's satisfying and good for their brains. You don't really *need* to learn much Chinese to travel in China, or even to live/work there.

 

That being said, one thing AI fundamentally can't do for you is be you; any conversation mediated through an AI is necessarily going to mean either a synthetic voice or you pronouncing some text on a screen, so even if the grammar is perfect, the other person is going to be well aware they're talking to a robot. So if you want to develop Chinese-language personal relationships, an AI is never going to replace the human element in those.

[realmayo]

On 5/30/2023 at 2:07 PM, mikelove said:

why does anybody bother to learn arithmetic when you can have a calculator do it for you?

Why does anybody learn to memorise how to write Chinese characters when you can have an app do it for you?

[Jan Finster]

On 5/30/2023 at 4:40 PM, realmayo said:

Why does anybody learn to memorise how to write Chinese characters when you can have an app do it for you?

 

That is quite on point. Unless you are forced to (HSK, school/uni in China), more and more people no longer bother to do so.

 

On 5/30/2023 at 3:07 PM, mikelove said:

going to mean either a synthetic voice or you pronouncing some text on a screen, so even if the grammar is perfect, the other person is going to be well aware they're talking to a robot. 

 

I have listened to nearly hundred hours of podcasts on AI and as far I understand, in the not very far future, AI will be able to mimic your voice. The other day we had a post on here linking to a video of an AI speaking like David Attenborough in German (David does not speak German). It was absolutely believable.

 

On 5/30/2023 at 9:11 AM, Lu said:

Google Translate is now 17 years old and it still isn't all that great for Chinese

 

I do not believe this is a good comparison. First, it does not matter, how old Googletranslator is. There have been recent breakthroughs in machine-learning that were not even conveiable 15 years ago. Also, according to my understanding an LLM generated its translation quite differently from Googletranslator. There may very well be some specialiced LLMs in the future that are particulary good at a certain topic, e.g. tanslating political topics or historical topics depending on which topics they are fed.  I have tried Deepl and ChatGPT on medical topcis German to English and v.v. and it is pretty, pretty good!

 

Just as a side note: I think GoogleTranslator has come a long way and is much better than 10 years ago. I remember when it would translate "Ich habe einen Kater namens Tom" to "I have a hangover called Tom" ...["Kater" in German can mean either tomcat or hangover]

 

[mikelove]

On 5/30/2023 at 2:09 PM, Jan Finster said:

AI will be able to mimic your voice

Sorry, that's not the issue, I'm talking about the fact that it's not actually coming out of your brain and your mouth. Having a conversation with somebody where they type something into a keyboard or say it in their native language and then it plays you a translation in their voice isn't really any different than the same thing with a computer voice; either way, you're not really talking to this person but rather are talking to some intermediary.

 

So it's a "robot" in the sense that there's obviously a computer in the middle of this, not because it sounds robotic.

[becky82]

On 5/31/2023 at 4:30 AM, mikelove said:

So it's a "robot" in the sense that there's obviously a computer in the middle of this, not because it sounds robotic.

 

Dodgy taxi driver: *switches note* 你交了5元，不是50元。

AI translator: You paid 5 yuan, not 50 yuan.

Me: [redacted]

AI translator: As an AI language model, I cannot accuse someone of being a total and utter [redacted].

[Jan Finster]

On 5/30/2023 at 10:30 PM, mikelove said:

either way, you're not really talking to this person but rather are talking to some intermediary.

 

I see. I agree. Nevertheless, this [talking to some intermediary] is essentially what is happening already when e.g. XiJinPing and Biden talk to each other via human interpreters. They hear the interpreter's voice, which is very different from the voice that is actually speaking. So  imagine, if AI could actually perfectly mimic Biden'd voice in Chinese (and XiJinPing's voice in English), this would probably actually be an improvement.

[Lu]

On 5/31/2023 at 5:45 AM, Jan Finster said:

Nevertheless, this [talking to some intermediary] is essentially what is happening already when e.g. Xi Jinping and Biden talk to each other via human interpreters. They hear the interpreter's voice, which is very different from the voice that is actually speaking. So imagine, if AI could actually perfectly mimic Biden's voice in Chinese (and Xi Jinping's voice in English), this would probably actually be an improvement.

Exactly, and they would probably have a better conversation if one of them could speak the other's language well enough. Or take a more common conversation: a doctor who only speaks language A, and a patient who speaks only language B. An interpreter is incredibly helpful there, but it would be even better if they could speak directly. More emotion and understanding is conveyed that way.

 

'Mimic Biden's voice', but Biden doesn't have a Chinese voice. Biden's way of speaking is for a very large part shaped by his culture. American people react to a certain inflection in a certain way, and on his way to success Biden has learned to adjust his way of speaking accordingly. Dutch or Chinese people react to that same inflection in a different way, even if they fully understand the language being spoken. Or look at the typical 'leader speak' that Chinese leaders develop: slow, with many pauses and 'hm's, specific emphasis. I'm sure AI could be made to mimic that very well, but it would not come across the same way at all. And that's not even going into body language and the reading of that body language.

 

I shudder at the thought of having AI interpret high-level political meetings. I know that human interpreters can make mistakes as well (since they are human), but the mistakes AI makes can lead to war or other horrible misunderstandings.

[realmayo]

On 5/31/2023 at 4:45 AM, Jan Finster said:

e.g. XiJinPing and Biden talk to each other via human interpreters.

 

Exactly: let's say Biden spoke HSK4 level Chinese. Is Xi going to want to struggle through that, or use an interpreter? Similarly if an HSK4 foreigner is in a cab with a chatty and friendly taxi driver, mightn't the taxi driver not prefer to have an adult conversation with next to no misunderstandings, mediated smoothly through an AI interpreter? This is my concern - because I used to absolutely love taxi rides in China, swapping smokes with the driver, slowly getting each other understood, even when I was at a very basic level. Perhaps that extremely cool aspect of learning a foreign language might diminish if you can just put a phone on the dashboard and it will translate like a good interpreter.

 

Don't get me wrong - I'm not saying this should happen, that it happening would be a good thing, or anything. Just that tech might suck some of the joy out of using a language at a far-from-fluent level.

 

If you're fluent then sure, 1000% better than tech. But will strangers still be patient enough to practise with you before you're fluent?

[realmayo]

 

On 5/27/2023 at 6:14 PM, realmayo said:

I assume this technology already exists.

 

Can't be far away: this is impressive simultaneous translation isn't it?

 

 

(from the second site I googled: https://sat-demo.github.io/  )

[Lu]

On 5/31/2023 at 11:40 AM, realmayo said:

But will strangers still be patient enough to practise with you before you're fluent?

Some will, some won't, but this is already the case. Some people will always be delighted with a foreigner trying to learn their language, and some people will always roll their eyes with the struggling foreigner who doesn't know their language already. For every two chatty and patient taxi drivers there is that person who insists on not understanding English while you speak Chinese to them.

 

That said, it sounds very plausible that AI will become a crutch the non-speaker will be expected to rely on instead of struggling through the conversation. Like the calculator that tourist shop attendants pull out when a foreigner wants to haggle. And that would really be a pity.

[Moshen]

Something that hasn't been emphasized yet is the requirement in a normal conversation to understand the intention behind the words said.  For example, haven't we all had the experience of chatting with a "bot" who completely misinterprets what you are really asking and gives a frustrating and mostly irrelevant answer?  

 

Add to that the cultural differences that come up in conversations between people from different cultures, and the potential for misunderstanding is huge when there is computer-mediated communication (whether with Google translate or AI).  The other day I was in a doctor's office with my China-born husband and when my husband was telling the story of his medical problem, the doctor asked a simple question, "So what happened?"  To me, this meant, "What was the outcome or diagnosis at that time?"  But my husband answered the question in a completely different way, that seemed to me irrelevant to the doctor's intention.  We argued about it afterwards, each of us convinced that we had the correct understanding of "So what happened?"  I'm pretty sure there was some kind of cultural difference operating there, though I can't say exactly what it was.

 

What I'm saying is that with automated translation, computers need to be culturally smart as well as linguistically smart, and that is very hard.  

 

Here's what ChatGPT said about this when I asked:

Quote

 

1. Cultural nuances and idiomatic expressions: Automated systems may struggle to capture the cultural nuances and idiomatic expressions present in the source language. These expressions often rely on context and may not have a direct equivalent in the target language, resulting in a loss of meaning or misinterpretation.

2. Wordplay and puns: Translating wordplay and puns can be particularly challenging for automated systems. These linguistic devices often rely on specific words, sounds, or double meanings that may not translate well into another language, leading to the loss of humor or intended effect.

3. Slang and colloquial language: Automated translation may struggle with accurately rendering slang or colloquial language. Such informal expressions can vary greatly across languages and regions, making it difficult for automated systems to provide accurate translations that capture the intended tone and meaning.

4. Names and cultural references: Automated systems may encounter difficulties with translating names, especially when they have cultural or historical significance. Proper names, place names, and references to specific cultural events or figures may require additional context or localization, which can be challenging for automated systems to handle accurately.

5. Technical jargon and domain-specific terminology: This includes technical jargon or domain-specific terminology, such as scientific or legal terms. These specialized terms may not have direct translations or equivalents in the target language, requiring manual intervention or domain-specific knowledge to ensure accurate translation.

 

Regenerate response

 

[mouse]

On 5/29/2023 at 3:30 PM, Jan Finster said:

I do not think speed is the issue here.

 

I was actually thinking of word order.

 

Translating a formal speech is one thing, but - and I may be too fixated on your babelfish comparison here - informal human conversations between equals is something else, and I think using AI to converse may well end up being doable, but ultimately superficial and not enjoyable. I would also add in passing that people still, despite the financial, time, and environmental costs, and despite the availability of videochat, fly around the world to do deals and attend meetings in person.

[realmayo]

On 5/31/2023 at 2:49 PM, Moshen said:

What I'm saying is that with automated translation, computers need to be culturally smart as well as linguistically smart, and that is very hard.  

Given that huamns aren't great at this - as your example with your husband demonstrates - I dont think it's necessary to hold AI to a higher standard than real people.

 

Personally I can easily imagine a world in which we all have our individual AIs which have got to know us and our subtleties of expression far better than any human interpreter could (apart from ones people are married to). Your AI will collaborate with the other person's AI to make sure that your quirky language use or cultural reference is translated into the other person's quirky linguistic or cultural referent context!

 

Somewhere I saw something about a dating app where the AI learns how individual users interact with humans: then it simulates users interacting with each other, 1000 online dates in a couple of minutes, and reports back on the best results to the users. A bit like this: https://youtu.be/It4Zbshd9uk?t=47

[mouse]

On 5/31/2023 at 3:35 PM, realmayo said:

Personally I can easily imagine a world in which we all have our individual AIs which have got to know us and our subtleties of expression far better than any human interpreter could (apart from ones people are married to). Your AI will collaborate with the other person's AI to make sure that your quirky language use or cultural reference is translated into the other person's quirky linguistic or cultural referent context!

 

 

Yes, ChatGPT's ability to build off previous interactions does suggest this will be doable.

[Jan Finster]

I am surprised there is so much opposition and negative sentiment towards AI and languages. Do you guys not see the upsides? I love traveling and having an AI assist me when talking to people in Ethiopia, Oman, Thailand, Myanmar, Borneo, etc is an absolute godsend and will surely positively improve the interaction I can have with locals...

[Moshen]

Quote

so much opposition and negative sentiment towards AI and languages. 

 

I think yuou've misinterpreted the comments here.

 

The question here was not whether there are some good uses for AI in communicating across language barriers but whether Ai would make language learning obsolete or unnecessary.  It's the latter question that brought out "opposition."