Yeah, my interest was piqued by the title but that demo video torpedoed it... Which is such a bummer because language learning is a perfect use case for LLMs
I'm curious how much effort goes into the management of these LLM backed language tools? I'm learning a less popular language (Dutch) and almost none of them support the it out of the box. Is this a marketing decision? What is the overhead to add support for a new language?
For those interested in this, there are a lot on the app stores these days that integrate this a bit better, i.e ask what you want to talk about first, and then you communicate for a short time with the AI about the topic.
It is rather for people who knows some basics, but would like to improve their language. It is necessary to be able to chat at least at some basic level.
I have a feeling that soon we will stop learning new languages to communicate with others because real-time AI translation will be so good, and we'll learn languages in a vacuum for intellectual amusement only, barely getting past the A2 level due to the lack of true human stimulation in the activity. This tool is certainly a step in that direction, and it strikes me as a great loss to humanity because we are losing our diversity. It just makes no sense to equalize all possible abilities over all human beings so we have nothing to offer each other.
Depends on the language and the reason to learn it. Right now its pretty discouraging. I pick up, say, enough Italian to get through a 2 week holiday, but all the people I interact with are under 40 and keen to practice their (excellent) English. English is so much the Common Tongue that I even hear the French and German visitors using it.
But: if I can get thru that, and start to engage with (say) Italy at a deeper level, away from tourist activities, or starting to engage with their literature, then learning the language will have a huge payoff. It's just a bit disheartening when you struggle to reach A2 or B1 but can't use it in the country because everyone switches to English. I guess my accent needs more work than I thought...
> But: if I can get thru that, and start to engage with (say) Italy at a deeper level
I was at a native+ level of English in high school according to an IELTS test I took for fun. During college my English improved further.
When I moved to USA … ho boy the difference between fluent English and culturally fluent idiomatic English is surprisingly vast. Much hilarity, frustration, and subtle misunderstandings ensued. It’s like an uncanny valley when your language is so good you start being judged as a native who’s a little weird and unsettling.
After 9 years of full immersion (I live here with nobody of my native language to speak with), you could say my English is near perfect. Hell I write and publish books in English! The next generation has started using new idioms I do not fully grok. What the hell is “based”?
Basically feels impossible for to ever fully catch up. Language moves too fast and your cultural background will never be same as the natives.
Edit: One super interesting aspect is that to speak fluent native English you have to do it wrong. Californians, for example, don’t use the plural contraction. You have to say “there’s many options” instead of “there are many options”. Otherwise you sound like a weirdo.
> Edit: One super interesting aspect is that to speak fluent native English you have to do it wrong. Californians, for example, don’t use the plural contraction. You have to say “there’s many options” instead of “there are many options”. Otherwise you sound like a weirdo.
You’re viewing it the wrong way around. L1 speakers almost never use formalized grammar rules (especially not in everyday speech). Instead, the rules are formulated post-hoc based on the way the language is actually spoken.
In a sense, there’s no ‘correct’ or ‘incorrect’ language use, only successful or unsuccessful attempts at communication.
For this reason, I find the topic of input-based language learning interesting. The basic idea is that the only true way to acquire a language is by getting a lot of input (i.e. immersion), not by studying grammar rules.
Not sure what you mean by this. This post-hoc formulation is how you define formal grammar rules, at least in a modern linguistic approach.
If the grammar rules you're referring to are nonsense prescriptivist things like "don't end a sentence with a preposition" that you get taught in school, then yeah you can completely disregard them.
It is possible to codify the actual grammar of a language as it is spoken, and reading these rules is very helpful for learning a language, but this kind of accurate description of grammar is something you're more likely to find in academic papers than school textbooks.
I’m referring to the way language students are taught, which is usually in a classroom setting involving lots of grammar drills.
As a personal example, when studying German in high school, I had to rotely memorize charts for article declension [1], prepositions [2], etc. Being able to regurgitate these charts helped me pass standardized tests, but they didn’t actually improve my ability to speak German.
Similarly, students of English are encouraged to painstakingly study rules such as those differentiating ‘I’, ‘my’, ‘me’, ‘mine’ and ‘myself’. [3] Personally, I learned English through immersion instead, acquiring these rules subconsciously rather than consciously studying them!
This is what input-based learning gets at. The theory goes that understanding input, and not conscious learning, is the only way to increase linguistic competence. [4]
> Californians, for example, don’t use the plural contraction. You have to say “there’s many options” instead of “there are many options”. Otherwise you sound like a weirdo.
Don't extrapolate too far from a small sample. I don't think anyone would recognize you as a non-Californian just for saying "there are".
Now, if you refer to highways US-101 or I-5 as anything other than "The 101" or "The 5", then you're clearly not from here!
Wikipedia does not back me up on this, however. Apparently the "The" names are SoCal-indicative. This has not been my experience here in NorCal.
FWIW, I learned the meaning of "based" once, but I found the explanation dissatisfying and promptly forgot it. :)
> Now, if you refer to highways US-101 or I-5 as anything other than "The 101" or "The 5", then you're clearly not from here!
Ah see here's where it gets interesting! Up here in SF, 101 and 5 are names. You say "On 101". It's the weirdos down there in LA who say "On the 101" :)
There needs to be a name for this phenomenon experienced by native English speakers abroad, who actually want to learn a foreign language.
I knew my Japanese was getting decent, many years ago, when my Japanese coworkers started replying to my Japanese in Japanese, instead of replying to my broken Japanese with broken English.
Sometimes I ask myself if WALLE is not that unrealistic of an dystopy, as we can already see that people are worse at remembering numbers in contrast to the time before smartphones, that save this information for you. Right now you can let ChatGPT do your homework without knowing anything about the subject. Where does this stop? Are we steering for a future where machines are doing everything for us and we will become dumber and dumber?
From the perspective of a hunter gatherer, we already live in this future. Our skills have nothing to do with theirs. From their perspective we are phenomenally dumb because we couldn't survive very long in the wild, and know very little of our natural surroundings.
In fact you can argue natural pressures have stopped us from evolving. We don't need our logical skills anymore to deduce the tracking and pathing of animals we hunt, or our memory to remember where the berries are.
i think what people memorize has been made optional, but that people still memorize a lot. sure not phone mumbers, but the amount of lore they can recite from some fandom seems to be of similar depth
There’s always something lost in translation, no matter how good the translator is, and the less the languages are related the more is lost. AI translation won’t change the motivation for language learning that much. It’s more the translator profession that will suffer (and already suffers).
Because forcing people to speak the ruling class's lingua franca has done so much good for the 500+ languages we already forced out of existence? If anything, finally having proper real-time translation would allow folks to keep speaking their native tongue instead of being forced to abandon it.
Languages with global speaker bases will be fine, but languages with tiny speaker groups would highly benefit from having an alternative to "stop speaking your own language and learn ours, or your government won't talk to you".
I have a feeling that soon we will have models which will predict the future better than humans and then the humans will stop predicting it and we will lose all insights into how to predict the future. at least the models will not make mistakes. /s
I was definitely hoping for more depth here.