AI & Language Learning (1)
You have to learn on your own because you're motivated to learn.
Hi there, Steve Kaufmann here, and today I want to talk about artificial intelligence
and it's application to language learning.
Remember, if you enjoy these videos, please subscribe.
Click on the bell for notifications and uh, please follow me on a podcast
service, especially if you're working on your English, you can carry the
audio with you and, and, um, I find that very effective in language learning.
So, artificial intelligence.
Um, you know, I use artificial, first of all, I should say that, uh,
October the eighth was my birthday.
Uh, my son and his family were over and I got this new golf shirt, so I'm happy.
77th birthday.
Anyway, artificial intelligence, so I use artificial intelligence, uh, in
the way I learn languages already, and I think we're just at the beginning of
what artificial intelligence can do.
I use it, for example, when I, uh, watch a television program on MTV Lebanon,
uh, or some other website in Persian and then I download the MP3 file and
then I put it on Happy Scribe, which is an automatic transcription service.
So that is artificial intelligence as I understand it.
I see artificial intelligence as using, uh, basically computing power to do
things much more quickly and much more efficiently than humans could do it.
And thereby giving humans the benefit of this tremendous ability to select and
to, uh, parse and to divide and, and all the different things that, uh, that, uh,
the computer can do to audio and to text.
Uh, and thereby make it easier for a learner to learn from that, uh, material.
So I put, I get these MP3 files, I get this trans, you know, transcribed.
And then I'm able, and maybe it's a type of, uh, artificial intelligence,
I'm able to import the audio and the text very easily into LingQ.
So then I have the ability to study this material on LingQ.
So that's a basic use of artificial intelligence.
I think every time I go to Google Translate to translate something,
the machine is translating it for me.
So in both cases, the computer is going at this large corpus of, uh, you know, audio
or text material that it's been through, and is able to draw certain conclusions
very, very quickly, much more quickly than a human could possibly dream of doing.
So you get an instant translation, which may or not, may not be as
accurate as what a human could do, but that accuracy is improving.
So, but, but where do I see auto, or at least the artificial intelligence going?
So I did some Googling, uh, of, you know, how people are using artificial
intelligence for language learning.
And of course a lot of it relates to the sort of teacher perspective.
So we will use artificial intelligence to provide feedback to the learner
because the teacher thinks that a major role for the teacher is to
provide feedback to the learner.
Personally, I'm not that interested in feedback.
Or artificial intelligence can be used to evaluate the learner, grade the
learner, maybe grade the learner in a way that's more private than or less
damaging to the learner's self-esteem.
Uh, or less time consuming for the teacher, but still evaluating the learner.
I'm also not very interested in being evaluated, so neither of those
two things really turn my crank.
Then they talk about, you know, coming up with textbooks that are sort of catered to
the learner's weaknesses or difficulties.
Now we're getting a little closer, but that's not how, how I would use it.
What I see artificial intelligence and, and what I think we're gonna try to
introduce more and more at LingQ is be because language earning is, is very
much an individual thing, a personal thing you can't, you can't expect people
to constantly correct you and guide you and give you feedback and evaluate you.
And that somehow that way you're gonna learn.
You have to learn on your own because you're motivated to learn
and therefore you wanna find content that's at your level.
That's of interest to you.
You know all the different ways in which you can define content
that's gonna help you learn.
If you identify a voice or type of voice that you like, maybe you can find
more material with that kind of voice.
For example, you can find more material uh, of a certain regional accent,
perhaps maybe I could, uh, I should be able to dictate to the using, you know,
voice commands, which is artificial intelligence and say find me something
that's 20 minutes long on the, uh, recent, uh, you know, demonstrations in Iran.
Maybe some discussion.
Maybe I could even say that I want the percentage of new words to be X, sort of
thing, might be more difficult because there will be some competing definitions.
Maybe if you want it too simple, it can't deal with more complicated
topics, but these are things that would have to be worked out.
Or I want a book, I want a book, an audio book and a and um, electronic
book but I don't want it too long.
And again, I want it on a certain subject, maybe Iranian
history or something like that.
Maybe such a thing doesn't exist, but the amount of time that I spend looking
for material, if that could be automated, so someone could automatically find
that for me, that would be amazing.
Now we also have text to speech, which is a form of artificial
intelligence, but most text to speech is still not so pleasant to listen to.
The voice sounds a bit artificial, but it's getting better all the time.
Is it possible that we would get to where, you know, you'd find a text
somewhere, something written in Wikipedia or whatever it might be, or a book
that doesn't have an audio version.
It's not an audio book, but the quality of text to speech would be so good that
would actually be pleasant to listen to.
A possible use of, um, artificial intelligence once you have material, say
that you've imported into LingQ, another problem I have with a, with a current
state of artificial intelligence is time stamping because when I study content at
LingQ, I like to study in sentence mode.
So I have a sense, and I would like to hear the audio just for that sentence.
But the timestamping is not accurate.
You know, I use, uh, Happy Scribe.
It says here you can timestamp.
And even within LingQ, we have an ability to, you know, re-timestamp certain texts.
Apparently it's better in other languages, but in Arabic and
versions it doesn't work very well.
So very accurately timestamped material, matching the sentences that I'm studying
in LingQ, that raises another issue if I get something automatically transcribed...
now again, the accuracy of that transcription is improving all the time.
So if I have audio material, I would like to read it, I go to
Happy Scribe, it's 90% accurate.
That's a problem when you're learning the language.
Another problem is there's no punctuation.
So therefore, some of the sentences that I would like to study in
sentence mode are in fact a paragraph because there was no punctuation.
So as all of this improves, I could maybe say, I want the sentences to
be between eight and 12 words long.
And so then, The text and the corresponding audio would be broken
into sentences that shouldn't be less than eight words long and shouldn't
be longer than 12 words long.
I'm just thinking outside the box here, I don't know to what extent that
is practical or not, but there's a real advantage in studying material
in what I call sentence mode, Uh, in terms of, you know, maybe reviewing
the new characters that were in that sentence, the ability to sort of
scramble the sentence and put it back together again.
The, uh, matching pairs exercise you can do with the three or four
words that were in that sentence that you're still trying to learn.
So if it were possible to find content that's of interest to
you, the total lesson shouldn't exceed, let's say 15 minutes.
Because you lose your, I don't know, you start to want to get, get away
from this text after about 30 minutes because it takes so long and, and yet
divided into sentences with the matching audio uh, that would be a wonderful
use of, uh, artificial intelligence.
Maybe I could say I'm looking for certain, um, vocabulary so I could write
something in English with the vocabulary that I want have it very accurately
translated into the target language.
Say, you know, Levantine Arabic, and then I would have very pleasant audio, you
know, text to speech in Levantine Arabic.
Is that ever gonna be possible?
I don't know.
Or is it always gonna be preferable to listen to an authentic, you know,
voice, this raises another issue.
When you look at the, uh, things that people recommend for, you know, potential
use for, um, artificial intelligence, they always have these language bots.
So you can talk to a robot and the robot answers you, but how meaningful is that?
I've never done that.
There's no question that if you have a, a genuine, spontaneous interaction
with someone in the language you're learning that is very high resonance.
You're struggling, you're clinging to every word, that's very high resonance.
If you have a lesson with a tutor, it's, it's still a human being.
It's, it's, it's better.
It's a little more authentic, but it's also a little artificial.
So not quite as good in my view, as that random, genuine encounter with someone.
So what about a chat bot?
Now what about if I'm talking to a machine that's answering me, I
don't know how much I'd like that.
Would I want that chat bot to correct me?
I don't even want my tutor to correct me.
I mean she doesn't, I, I, it's not that I would react
negatively, but I'm not interested.
I'm having a conversation.
I'm enjoying our conversation.
I don't necessarily wanna be corrected.
So if you have a chat bot, a bot correcting you, how useful is that?
Again, I don't know.
I'm just throwing stuff out there.
Potential uses of AI, already it has changed the way we learn.
I use text to speech all the time.
I use Google Translate all the time.
I use these automatic transcription services when I'm in YouTube
and I'm looking at a, a video on whatever the subject might
be, Uh, something from Lebanon.
It's gonna suggest a whole bunch of other, uh, related type videos.
So is all of that gonna get better?
So it's becomes easier and easier to find content of interest and then move
from there to a related content of interest, all of which helps you sort
of zoom in on a particular range of, of vocabulary, all of which makes the job
of the teacher easier because the teacher doesn't have to prepare the material.
You don't have to buy a $200 textbook, such as I saw when
I visited Stanford University.
Language textbooks selling for $200.
You can get access to all of this material.
And, uh, and even in terms of, uh, the grammar, I mean, again, AI can help
you identify this particular word.
What is, uh, you know, the conjugation of that verb?
And, and, uh, so forth.
You might be able to ask for material which focuses in on the subjunctive mood
or the instrumental case or anything in different languages, and get a whole
bunch of examples of that particular structure, that pattern in the language
for you to listen to and read again.
It seems to me the potential, the opportunities there are endless in