March 26, 2025 • 37 mins
It feels like everyone is talking about artificial intelligence and how it's taking over almost every industry. But what about language learning? Join Marina and Jeremiah as they discuss the benefits, shortfalls, and techniques when it comes to using this powerful tool to help you learn a language.
Sources:
Schmidt, T. & Strassner, T. (2022). Artificial Intelligence in Foreign Language Learning and Teaching. Link
Son, J., Ružić, N. & Philpott, A. (2023). Artificial intelligence technologies and applications for language learning and teaching. Link
Woo, L. & Choi, H. (2021). Systematic Review for AI-based Language Learning Tools. Link
New episodes every month! Transcripts available at howtolanguagepod.com. To stay in the loop, follow us on Instagram @howtolang
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Marina (00:02):
Hello and welcome to How To Language a podcast about
language learning.
I'm Marina,
Jeremiah (00:07):
and I'm Jeremiah,
Marina (00:08):
We know you're probably sick of hearing about AI in the
news, but today we're divinginto the world of AI and
language learning how artificialintelligence is changing the way
we learn languages, what it getsright, what it absolutely
doesn't, and how to use it as atool without letting it replace
the most important parts oflearning a language.
Jeremiah (00:28):
Yeah.
And, don't worry, we didn't useAI to write this intro.
Mostly
Marina (00:35):
So we're talking about AI in language learning, but I
think before we dive too deep,we should explain what we mean
when we say ai.
'cause that is kind of a scarybuzzword that is thrown around a
lot.
But I think we should clarifywhat exactly we mean.
Jeremiah (00:53):
Yeah, for one thing, because when we say it, we mean
multiple things.
Kind of interchangeably, eventhough they're not
interchangeable.
And for two, I think, you know,people have been kind of
anticipating AI for probablysince 2001, a space oddity.
Odyssey, sorry.
I was getting my, my pop culturemixed up also.
(01:16):
It didn't come out in 2001.
That's just what it's called.
When did it come out?
1941.
I dunno.
I bring it up because wasn'tthere famously an AI in it?
Marina (01:27):
Yeah.
Hal,
Jeremiah (01:28):
how
Marina (01:28):
the ai Yeah.
Jeremiah (01:30):
I've never actually seen that movie.
Was Hal good or bad?
I
Marina (01:32):
think the whole point was that he was bad.
Yeah.
Jeremiah (01:34):
Usually the AI is bad.
Like if you've played PortalUhhuh, if you've seen Wally, if
you've seen W Yeah, that was a
Marina (01:42):
pretty direct reference.
Perfect
Jeremiah (01:43):
example.
If you've seen, um.
iBot.
The whole point of that moviewas that AI is bad.
But in all of those cases, theAI at least is, it's implied
that humans have figured out howto actually make an artificial.
Intelligence, it's in the name.
And one important thing to pointout is that in our timeline,
(02:05):
that is not what happened.
Like none of this is actually anartificial intelligence.
None of it is like a digitalbrain that does all the same
things your brain does.
We'll get into it, but like,it's fundamentally not what it
sounds like.
So I'll hand it back over to youto talk about what it actually
is mm-hmm.
At this stage in history anyway.
Marina (02:23):
So what we have currently, as far as I
understand it, is narrow ai,which means basically that it's
an artificial intelligence thatis designed and programmed to.
Do one specific task.
It doesn't have the greaterthinking capabilities that are
(02:45):
portrayed in movies like that.
Exactly.
Jeremiah (02:48):
Like you couldn't, put chat.
You couldn't download chat GPTinto like a robot body and then
have it discover fire orsomething.
Marina (02:56):
Right.
Like.
Chat t's only purpose that itwas designed for was to mimic
human language.
Jeremiah (03:04):
Exactly.
Yeah.
if you've ever played a videogame, all the other people in
the game that aren't real, allthe NPCs, they are all run by a
very narrow ai.
Like if you play Skyrim andyou're fighting bandits, that AI
has one goal, which is to killyou.
So you know, a perfect example.
It could be literally one onespecific task.
Yeah.
And it counts as AI apparently.
Marina (03:25):
Exactly.
And some of these AI tools aredesigned with language teaching
as their main task.
Some of them.
Can do that as a part of theirexisting tasks.
So, there's a few differentkinds of AI that we see in the
language learning landscape.
There's translation AI toolslike Google, Google Translate,
(03:49):
and Deep LI don't know if that'show you're supposed to say it.
If it's like de I don't thinkit's de, I dunno, there's so
many dumb names like thatnowadays.
There are a lot of dumb names.
You can never be sure.
So they like
Jeremiah (04:03):
chat.
PPT, sorry, I'm cutting thatout.
No, no, you can't cut that.
Okay.
Keep going.
Marina (04:11):
A lot of people don't think about Google translate as
ai, but it technically is andit's task is translation.
There's of course things likechatbots and natural language
processors.
So that's what people usuallyare talking about nowadays when
they talk about ai.
This is like chat GPT.
(04:31):
deep seek.
Grock, Elon Musk's ai.
Apparently.
There's writing evaluation toolslike Grammarly.
Those are super popular for.
Editing essays and giving youtips on writing and revision.
And then there's also thingslike what are called ITSs or
(04:52):
intelligent tutoring systems.
There's a lot of these fordifferent subjects, but the most
well-known one in languagelearning is Duolingo, which
isn't entirely ai, but it hasai.
Features in it that kind of helpassess your language learning
progress and adapt their lessonsbased on how you're doing.
Jeremiah (05:12):
Yeah.
It seems like what all thesekind of have in common, if you
think about it and part whatmakes them ai other than, oh,
they're programed to do a task,which is a lot of things that
don't count as AI is they all onsome level are like capable of
independently making decisions.
Even if it's really smalldecisions, like for Google
Translate, for example, it'sprogrammed to be able to decide.
(05:34):
How to word a sentence based oncriteria that are not obvious.
And for like Duolingo, I don'tknow, it decides like, when to
annoy you on your phone withyour push notifications.
Marina (05:47):
Mm-hmm.
Jeremiah (05:47):
And maybe we'll get to this, but speaking of chat bots,
they've started actuallyincorporating those in the
intelligent tutoring systemsnow.
Like memorize has little chat.
AI buddies that you can talk to.
I've tried the Chinese one andthe problem is it tries to, it
instantly uses words that Idon't know.
The problem with Chinese is youcan't just look at the words and
(06:08):
figure it out.
Unless you've seen the wordsbefore, it's like, might as well
be Greek for all, you know?
Marina (06:12):
So, Jeremiah, have you used ai?
It sounds like you have.
Jeremiah (06:18):
You know, I used to use Chate a lot and then I got
sick of paying for it when Irealized that Deep Seek can do
most of the same stuff for free.
So now I use deep seek if I havequestions.
And what else?
I had something else.
Speaking, yeah, experiences withai, actually when I was taking
old Norse in grad school.
(06:39):
For funsies.
I got kind of overwhelmed witheverything that I had going on
and I was like, I don't reallyactually care about this.
Like, I like going into class,but I don't, I'm not gonna like
stick with it enough to botheractually learning anything Oof.
Which is sad.
I really liked the instructor.
She was great.
But, so what I would do is Iwould just like feed Chacha Bee
(07:03):
my old Norse homework, mytranslation, what it came back
with was actually really good.
I got really good grades on it.
Marina (07:12):
That's exactly what you're not supposed to do.
I
Jeremiah (07:15):
know.
And here's my disclaimer forpeople listening.
If it's something you care aboutand you actually want to get
better at, don't use AI likethat.
You can still use it, but don'tjust have it do your work for
you.
That goes for anything.
But if you're taking one quarterof old Norse and you don't plan
on taking a second quarter orever using it.
You know?
Sure.
(07:35):
If you don't wanna fail, thenyou can use some ai.
I was surprised at how accurateit was though.
It was crazy even back then.
This was 2023, I think.
Marina (07:45):
Yeah.
Wow.
You just admitted to academicfraud.
Jeremiah (07:51):
I did what?
Marina (07:56):
All right, well, don't do that.
Jeremiah (07:58):
Don't do that.
Marina (08:01):
Okay.
So.
Ai, as a language learning toolis under this umbrella in second
language acquisition studiescalled Computer Assisted
Language Learning, which is, youknow, shortened to call,
(08:23):
everything's
Jeremiah (08:23):
an acronym these days.
I
Marina (08:24):
know, and it refers to.
The use of computers andtechnology to aid in the
language learning process,
Jeremiah (08:35):
which is so funny.
It's, it sounds like somethingthat, um.
Was like a research area in like2003 when computers were still
new and they were like, did youknow you could use computers to
learn a language?
And everybody was like, what?
That's crazy.
Marina (08:52):
I mean, yeah, that is kind of where it started.
Like, like CD ROMs that, oh mygod.
Like,
Jeremiah (08:57):
I don't even know what that is.
Marina (09:01):
So yeah.
But no, it really, it startedlike back then and then, you
know, it's.
Persisted all the way up untilnow with AI and things like
voice recognition software,things like that.
And so there's a pretty longhistory of research in Call that
has shown that technology can infact enhance language learning,
(09:24):
in several ways, which we'll getto shortly.
And that language learnersgenerally report positive
experiences and outcomes whenusing these tools.
So they are a good thing.
It's not like they hinderlanguage learning at all.
They tend to have the oppositeeffect.
Jeremiah (09:42):
Yeah.
It's not cheating.
Just like, you know, the firstfarmer who invented a plow
wasn't cheating.
Right?
Marina (09:48):
Yeah, exactly.
It's just another tool.
But you know, call hastraditionally been seen as a
supplementary tool, so it's nota replacement for real world
practice or language teachers ortextbooks or anything like that.
AI is just the latest evolutionof call with its own benefits
(10:12):
and limitations, which we'regoing to get into.
We'll talk about the benefits ofusing AI specifically in
language learning.
Things to look out for, thingsthat you're missing out on.
And we'll also go into some do'sand don'ts if you choose to use
AI as a supplementary tool.
So the benefits of using ai, Ithink the first one that comes
(10:36):
to mind and that's the the firstreason that a learner turns to
AI in the first place is thatit's an on demand tool.
It's almost like an on-demandlanguage partner.
Instead of going out and seekinga tutor.
Or visiting your teacher'soffice hours.
Um, you have immediate access toa partner to practice speaking,
(11:00):
writing and comprehensionwhenever you want, wherever you
are, which eliminates the needfor those human partners.
In theory it shouldn't replacehuman partners, but that is what
you can use it for in a pitch.
Similarly, you get immediateassessment on the output that
(11:22):
you generate, whether it'sspeaking to chat GPT or writing
your translation in Duolingo.
It gives you immediatecorrection and hopefully
explanation.
Addresses your mistakes in realtime, which is really a valuable
learning experience.
'cause making an error doesn'treally help you very much unless
(11:44):
you understand why it was anerror and how to correct the
mistake.
But in a classroom environment,you.
Can potentially receiveimmediate correction and
feedback from your teacher orfrom a tutor.
But that can also be scary.
It can be embarrassing.
It can be demoralizing to getthat from an authority figure,
(12:07):
somebody that you trust orrespect.
And so, you know, but AI is justan ai.
You can't really get yourfeelings hurt by it.
And so in that sense, it's avery low stakes language
partner.
You can talk to it or write toit and feel that you're not
going to be judged by them.
(12:28):
Many learners won't feel asanxious when speaking in their
target language.
It's very much a no judgmentzone, which definitely is a huge
benefit, if, especially if youstruggle with anxiety when it
comes to speaking your targetlanguage.
And I think the last mainbenefit of losing AI is that it
(12:49):
can be very good at adaptiveteaching, which means like it
can tailor its lessons or its.
Responses to be specific to yourneeds and your interests.
You can have a conversation withchat GPT about photography, if
(13:12):
that's your interest, in yourtarget language and be able to
talk about it at length, whereasyour teacher or tutor.
Might not be able to do that.
Um, they might not be able todevote that much time to it.
They might not know enough to beable to talk about it at length.
And so if that's something thatyou wanna do, that's something
that AI can provide.
(13:34):
It can also adapt to your level.
It can kind of get a feel forwhat level you're at based on
your output and adjust.
Its and adjust to, to that levelto make it more comprehensible.
Basically it can adapt to yourlevel and your learning styles.
(13:55):
And in fact, I think this isarguably the largest advantage
that AI has over traditionallanguage teachers and tools.
This is one of the reasons whysome teachers fear for the
future of the profession,because of the threat of ai.
I think that's not going tohappen, but it is something to
(14:18):
consider, because.
Teachers and other traditionalmethods like textbooks have to
cater to large swaths oflearners.
If you're teaching in aclassroom full of lots of
different people who are atdifferent levels and different,
learning styles, you can't tutoreach student individually.
(14:40):
So you have to kind of zoom outand do what's best for
everybody.
Now good teachers, part of yourjob is to do your best to
provide individual support andtake into account the individual
needs of your students.
But that can be really difficultdepending on the context that
you're in.
Jeremiah (14:58):
Mm-hmm.
A student or a,
Marina (15:00):
a teacher who has 10 students in her classroom,
that's a lot easier.
But if you have a multilevelclassroom full of like 40
students, like.
Good luck.
Yeah, that's really hard.
So, they can do their best, butit doesn't top an individualized
session with a, with a tutor,um, or in this case an AI tutor.
(15:26):
Because it can, you know,theoretically it can pick up on
what you're struggling with.
It can pick up on how you learnbest and then immediately apply
that to enhance the overallexperience.
The question though is whetheror not AI is capable of picking
up on those things and applyingthem effectively.
Jeremiah (15:48):
Yeah.
'cause you know that, that, thatis one of the consequences of
the fact that the AI is not likea functioning digital brain like
yours.
Like, like we have brains,right.
It's more like, a prettysophisticated, like pre
predictive text generator.
So the extent to which it cannotice things in the same way
(16:09):
that you and I can notice thingsis pretty, I would, I would say
dubious.
But on the other hand, you cantell it things like if you know,
oh, you know, this is what Iwanna work on, or whatever.
Ai, like chat bots in particularseem to do really well when you
give them a lot to work with.
Like, the more details youprovide, the more detailed their
(16:31):
response can be, becauseotherwise they're just working
with like the SumTotal of all ofthe training data that they
have.
But I think that segues uspretty nicely into, I won't say
the drawbacks of ai, and usingAI language tools, but more like
what you're missing out on, Iguess.
And I guess to start.
Maybe this is a drawback or moreof a consideration, I guess,
(16:53):
because some listeners mightjust have a really strong,
ethical, issue with even usingai.
In particular with usingchatbots, you know, whether
because they steal data orbecause they are hugely energy
intensive and those are fair.
So if you don't wanna use AIlanguage tools, I don't blame
(17:15):
you.
You're not gonna really missanything essential that you
need.
Whatever you're already doing isgonna be just fine.
However, if this is of interestto you, then by all means,
listen on.
So if you, if you are using AIlanguage tools, here's some
things you might be missing outon.
One thing I will say is thatsome AI tools like the Chatbots
(17:36):
have pretty advanced speakingand listening features, and in
addition to just writing, for awhile it was just writing, you
know, you would write to it andit would write back to you.
Kinda like, do you rememberClever Bot?
Marina (17:49):
No
Jeremiah (17:49):
way back in the day.
I don't know how that worked.
That was like a really,minimally intelligent, I guess
chat bot back in the day.
Back then that was just, thatblew our minds.
But you couldn't talk to it now,especially with chat GPT for
example, you can carry on a fullconversation with it, and it
almost feels exactly like you'retalking to another human.
(18:11):
And again, that's because it'snumber one job before anything
else is to convince you thatyou're talking to a person, not
explicitly, like if you ask it,are you a person, it'll say no.
But other than that, it's meantto create the illusion that
you're talking to a person.
So sometimes the AI model, youknow, the Shap PPT, will, even
(18:32):
include what we callparalinguistic features.
So those are things that are notwords, you know, they're not
part of the language itself, butthey're kind of things that go
around, the words that help givethe meaning, you know, the
expression that people say alot.
That 90% of communication isnonverbal, which is nonsense.
(18:54):
I mean, it's not really true.
But the kernel of truth to it isthat there's a lot of
communication that happensthrough body language and also
just like the things in speechin particular that are not
language.
So the pauses, like if you takean extra long pause between.
(19:15):
Your words to collect yourthoughts or, and, and, you know,
and these are things that, thatthe AI can do, right?
The, the chat, GPT, when youtalk to it, sometimes it'll
chuckle, which is kind ofcreepy.
That's creepy.
It'll pause.
Um, sometimes it'll say,
Marina (19:30):
Hmm,
Jeremiah (19:31):
or even clear its throat, which is really
unsettling.
However, it is important toremember that, ultimately it is
not another person.
So when you talk to a realperson, it's what it is, is it's
a relational process of give andtake where you are, you're
constantly listening andthinking and producing at the
(19:51):
same time in tandem with theother person who's doing the
exact same thing.
You know, it's like a weirdbalance where you ever find
yourself, like if you're in aconversation and you realize,
oh, I was just thinking aboutwhat I was gonna say next, and I
forgot to listen to what theywere saying, you know, that's a
constant, um.
It's a constant balance that youdo when you're in a conversation
with somebody.
(20:12):
And an AI does not have to dothat.
It's not thinking and it's noteven really creating what it's
doing is recombining contentfrom, like, the source is, it's
some total of training data,most of which it just kind of
stole.
So in other words, it'sessentially, like I said, a
really sophisticated predictivetext generator.
(20:34):
Exactly like the one on yourphone when you're typing and it
suggests the next word for you.
It provides a compellingillusion or like a facade of an
interaction, but it is just thatit's an illusion.
All of this is to say that whatyou're missing out on is the
real human interaction, which isone of the primary, like fuel
sources for language learning.
You can think of it kinda likethe difference between a pilot,
(20:56):
you know, when they're firstlearning how to fly and they go
in the simulator.
You ever see that?
They have like a flightsimulator.
It's really sophisticated.
Have I ever
Marina (21:03):
seen that?
Yeah.
If you
Jeremiah (21:04):
go to a museum, you know what, if we go to the
museum a flight, I bet we can.
I bet there's a flight simulatorthat you can use.
Marina (21:09):
That'd
Jeremiah (21:09):
be fun.
But yeah, they get in a flightsimulator, it's like a video
game, but it's reallysophisticated and it has all
those, all the right buttons andstuff.
Mm-hmm.
but that's what they do.
And you know, they do that totrain.
But if you were to just take apilot who has only flown in a
flight simulator and then putthem in charge of a commercial
(21:30):
airliner and told all thepassengers that they would freak
out, they'd be like, what do youmean?
This pilot has never flown areal plane.
So, you know, there's, what I'mgetting at is just like there's
a difference between learninghow to fly in a simulator versus
flying a real plane.
There's a really importantdifference between talking to a
simulated person and talking toa real person.
(21:50):
You can learn a lot from thesimulator, but the real
experience is gonna present youwith challenges, obstacles, and
problems that are emergent.
Which means they kind of justcome as if from nowhere.
They're not pre-programmed orwhatever, and unique.
Like problems that other peoplemight have faced, similar
problems, you might have facedsimilar problems, but ultimately
(22:10):
the conditions are unique toyour situation.
And so it's not so much that youlearn how to handle everything
that might come up.
It's more that you learn how tonegotiate obstacles kind of in
QuickTime as they come up.
And that is one of the mostimportant language skills that
doesn't really get talked aboutsuper explicitly, but we call it
(22:32):
negotiation for meaning.
I think we've talked about itbefore.
We'll probably have a wholeepisode about it someday.
That's an important skill andyou mostly can't learn it
without using the language inpractice, in real life
conversations.
Another gap in AI languagelearning is that real human
beings, when you talk to them.
They have specific backgrounds,they have memories, they have
(22:54):
cultural heritages, and theyhave finite linguistic resources
that are based on theirexperience in the lifetime that
they learned their language,right?
Like they don't know every word,or they might know a word, but
forget what it means.
Or they might have an idea, butnot sure what the word is for
it.
AI doesn't have that.
People don't always say whatthey mean.
(23:15):
Sometimes people say theopposite of what they mean.
That's where a lot of thoseparalinguistic features come up.
Like that's part of how you knowwhen someone's being sarcastic.
Sometimes they forget the wordthey're looking for or they have
difficulty explaining a concept,and all of that is stuff you
have to contend with when youtalk to someone.
People also have wants andneeds.
The person you're talking tomight be in a hurry or they
(23:37):
might be tired or they might befrustrated.
And all of these conditions arewhat makes every interaction in
your target language a uniquelearning experience.
And AI has no single culturalreservoir that it draws on.
It can't forget words or be in ahurry.
It doesn't have wants or needs.
So when you're talking to an ai,you'll never be presented with
(23:57):
the challenge of communicatingacross a cultural barrier or
having to be more concise orspeak more quickly, in order to
hold the attention of someonewho is in a rush or having to
ask clarifying questions to makesense of what someone's trying
to tell you when they aren'texplaining it very well.
Like these are all experiencesthat we've all had when talking
to people, even in our firstlanguages.
(24:19):
And AI really can't replicatethat.
Because that's not what it'smeant to do.
Lastly, some AI is good atmimicking paralinguistic
features, like I mentioned, solike a chuckle or taking a
breath.
It's kind of creepy, but itcan't replicate body language
because it doesn't have a body.
And so it's not very good.
It's also not very good atsubtext, which is another key
(24:41):
ingredient in most humancommunication.
So these are all some thingsthat AI just can't do.
And if you never go outside andyou never talk to someone in
your target language, just likewe always say, you have to go
outside, um, you'll be missingout on a lot of these things if
you just talk to Chad gt.
Marina (24:59):
So.
That was a lot of information,but, we'll kind of summarize it
here at the end with some do'sand don'ts for using AI for
language learning.
Definitely do ask it to breakdown difficult grammar points
into simple language or ask itto rephrase an unclear
explanation in like yourtextbook or something your
(25:21):
teacher said, this is one ofAI's strengths is that it's
really good at rephrasing andexplaining things in different
ways that might make more senseto you, especially if your home
base like a textbook is not veryhelpful.
You can
Jeremiah (25:37):
also ask it why a word or a structure in your target
language is the way it is.
As you know, like me, I likelooking up the etymology of
words in my target languagebecause I find that helpful for
connecting the dots betweenwords and kind of seeing the
matrix.
So if that's something that youfind helpful, AI can be helpful
with that.
However, you should not, assumeeverything that it says is true
(26:01):
if it's important information.
Something that you're gonna belike quizzed on later.
Uh, or especially if the AI'sanswer feels off, then
definitely double check what ittells you.
Marina (26:12):
I feel like AI would be particularly unreliable when it
comes to etymology.
I feel like it would be prettyprone to just making things up.
Jeremiah (26:21):
Yeah.
Or, or
Marina (26:22):
rely on folk etymologies,
Jeremiah (26:24):
potentially.
Yeah.
Because it gets all, most of itsdata from what the internet.
Marina (26:28):
I think so.
Yeah.
Jeremiah (26:29):
So theoretically.
In its data banks.
There's just a ton of Redditexplanations of things like, oh,
the, um, the etymology of the Fword is fornicating under
consent of the king.
I would not be surprised if, ifChad PT has confidently told
someone that
Marina (26:47):
Mm, yeah, it's, it's probably not going to the Oxford
English dictionary as its firstetymological source I'm sure it
Jeremiah (26:55):
has.
Reddit in its data, but youknow,
Marina (27:00):
yeah, I definitely don't take everything that it says as
fact.
Like we were talking aboutearlier, definitely do use it to
give you feedback on yourgrammar, your word choice, et
cetera.
It's gonna be really good atgiving personalized feedback
based on what you say or write,but don't rely on it to
translate everything, just likeyou wouldn't trust everything
(27:22):
Google Translates says.
There might be some translationtools that are.
Quote, unquote better thanothers.
Like deep L is gonna be betterthan Google, Google Translate,
for example.
But that doesn't mean that it'sinfallible, and it doesn't mean
that they don't have theirlimitations.
So, you know, I don't think thatwe'll be saying goodbye to human
translators anytime soon.
(27:44):
You always wanna double check,double check what it gives you
if you're using it for that.
Jeremiah (27:49):
Yeah, absolutely.
Also remember that, an importantthing with any language is that
it's a living thing unlessyou're studying a dead language.
It is a living unevenphenomenon.
And what I mean by that isthere's a whole corpus of
colloquialisms and idioms andslang terms that.
(28:16):
Are one temp temporallyconstrained.
And what I mean by that is likepeople weren't saying lock in or
crash out before, the last year,I think.
And you know, that's in English,but they're also often
geographically constrained,right?
Sometimes it'll be the sameword, but spelled differently in
different places.
(28:37):
Or mean different things indifferent places.
So what I'm getting at is,remember that AI does not
necessarily know everycolloquialism, every idiom, or
every slang term, especially ifit is particular to a certain
dialect or a geographical area.
Marina (28:56):
And if you do ask it to use.
Slang and colloquialisms from aparticular region of the world,
it might rely on stereotypes toachieve that, which isn't ideal
either.
Jeremiah (29:09):
Yeah, good point.
Like if you say, oh, speak Chileand Spanish.
Mm-hmm.
I feel like Chile and Spanish isun irrationally.
Look down on for some reason.
Marina (29:19):
Yeah, right.
Exactly.
So we don't wanna do thateither.
'cause it is just getting itsinformation from humans and
humans are fallible and havebias.
Mm-hmm.
And we don't want that.
So this last thing, in ourresearch for this episode, I
found, an article written bySchmidt and Strasser, Schmidt
(29:40):
and Rasser in, 2022 calledArtificial Intelligence in
Foreign Language Learning andTeaching a Call.
God for intelligent practice.
Jeremiah (29:53):
Oh no.
And call is like all capsbecause it's a, it's a pun.
Okay.
Anyway, off to a great start.
Marina (30:01):
At the very end of this article, they, give this utopian
their words vision of what a,computer assisted foreign
language classroom might looklike in 2040.
And so we have an excerpt here.
I shortened it'cause it waspretty long, so.
Mm-hmm.
I just, just keep that in mind.
(30:22):
It's not a word for word.
But
Jeremiah (30:26):
yeah.
And it also just glancing overit, it looks like there's some,
academicy jargon in here.
Mm-hmm.
Not, not crazy, but a littlebit.
Yeah.
How about I, I'll just go aheadand read it out loud because
this looks pretty interesting.
Cool.
I love eviscerating academicjargony, writing because I
always find it so.
(30:47):
You silly.
All right, here we go.
It says in 2040, the foreignlanguage classroom will be an
evidence-based, Hey, that's us.
We're evidence-based and we'rejust based, evidence-based
blended learning environmentwith bring your own device
solutions in all classrooms andsubjects.
(31:10):
Is everybody gonna be able toafford those devices?
I don't know.
Anyway, in 2040, the printedtextbook will have become an
interactive, multimedia,adaptive learning and practice
environment that is perfectlyadapted to the needs of both
face-to-face teaching and thephases of individual practice
and self-directed learning.
(31:31):
Okay.
The digitally enhanced classroomof the year 2040 will help to
diagnose needs and learningprogress, providing direct
access to differentiated andneeds-based support services.
The classroom of the future willcleverly combine the advantages
of digital learning with provencomputer free methods.
(31:51):
Content and tasks forface-to-face teaching, which
will remain indispensable andhighly significant for
successful learning.
And even in 2040, we will stillhave and need teachers,
well-trained data literateteachers who are competent,
critical, and reflective intheir use of media and
technology support, and who willuse and who use empirically
(32:13):
established digital scenarios,uh, that are based on valuable.
Learning content and are linkedto meaningfully constructed
challenging learning tasks andexercise opportunities designed
to support individual learning.
Oh boy.
Okay.
What, what are your thoughts?
Right off, right off the bat.
Marina (32:32):
That last one was one sentence.
Oh my.
Jeremiah (32:34):
That whole thing.
Wow.
Okay.
Marina (32:37):
Um, it is very jargony.
But you know, in the context ofthe rest of the article, it
makes callbacks.
And makes more sense,
Jeremiah (32:44):
makes callbacks.
Oh God.
Marina (32:49):
Um, I think it's basically saying that the goal
here is an adaptive learningenvironment that still makes
good use of quote unquotecomputer free methods.
And still necessitates teachers.
(33:10):
So, you know, I think that'swhat they mean when they say an
evidence-based blended learningenvironment, for all classrooms
and subjects.
And so I think that that'sachievable for sure.
I think we're already kind of onour way there.
You know, it'll.
Be nice to have, interactive,multimedia, adaptive learning
(33:32):
for everybody.
That might be kinda tough by2040.
Um, Do you have any thoughts?
Jeremiah (33:41):
Yeah, I mean, I'm just thinking about do you remember.
In, I don't know if Japanese hadthis, but Spanish had this, when
I was taking Spanish inundergrad, whatever textbook we
had had its own like proprietaryhomework website Yeah.
That you needed a code to getinto, and that's where you
(34:02):
submitted your homework and likea lot of the class was just on
there.
From what I saw that did notchange one iota in my entire
college career.
I don't think it's any differentnow.
I feel like the textbookcompanies still have a monopoly
on, that, homework site.
When I think of actuallyexisting digital.
(34:25):
Whatever in the languageclassroom.
That's kind of what I think of.
Marina (34:30):
But that's not necessarily AI either.
No, but I mean, that's just adigitally enhanced classroom.
Jeremiah (34:36):
That's true.
I'm just, I'm just basing thisoff of what this thing says.
cause this doesn't talk about AIthat much, this little excerpt.
Marina (34:44):
No, it really doesn't.
In fact.
Jeremiah (34:48):
Which I think kind of gets to a point that I'll make,
which is that actually that youmade a while ago, which is that
this whole AI craze is just likeone more step in a long standing
trend in language learning andin learning in general to
incorporate, you know, quoteunquote technology, which is
(35:10):
funny because it's always been.
It's always incorporatedtechnology.
I think they just mean digitaltechnology.
So in that sense, AI isn't supergroundbreaking.
It's just like kind ofqualitatively differentiated by
how weird and futuristic it is.
Marina (35:27):
Yeah.
The full, the full section ofthe article does mention AI
looks like twice
Jeremiah (35:33):
Mm.
Marina (35:34):
For example, quote, the learning platform deployed in
2040 is thus an AI based digitallearning support system.
A resource, a tool for studentsand teachers, which will be used
to add value to learn to thelearning process.
Hmm.
And then later they say, quote,a sustainable approach to make
teachers become acquainted withAI technology within a rather
non technocratic narrative,would probably, et cetera, et
(35:57):
cetera.
but they do mention ai, butreally not that much for an
article that's entirely aboutai.
Jeremiah (36:03):
Yeah.
Marina (36:03):
But I think, yeah, I think that that does speak to
the fact that AI is just onesmall component of a digitally
enhanced or call approach tolanguage learning.
You know, it's sort of thebuzzword now, but you know.
It really is just one component.
Jeremiah (36:21):
Yeah, I guess I really don't have that many gripes with
this article.
Then other than that, they'rekind of just yapping about, you
know, basically what we have nowexcept without, weird.
Monopolist, textbook companiesand technical difficulties,
which I feel like is one of themain issues with any digital
(36:44):
anything in the classroom, isthat it's always a pain to use.
Like canvas.
Oh my god.
Mm-hmm.
Um,
Marina (36:52):
yeah, I think, different platforms and tools have
different strengths and I don'tthink that there's really
anything inherently wrong with.
Combining them?
No,
Jeremiah (37:02):
no.
Marina (37:02):
You know, instead of just relying on the digital
textbook that gives you shortfilms to watch and makes you
order these words in the rightorder to make a sentence for
your homework kind of thing.
I think that if we could makeuse of all the best parts of
technology and the best parts ofAI for the particular learner,
Jeremiah (37:24):
that would be ideal.
Well, thank you for tuning intothis episode of How to Language.
We hope that you're walking awayfeeling better equipped to use
AI thoughtfully in your languagelearning if you want.
And if you like what we do, youcan rate and review us on your
preferred podcast app and youcan follow us on Instagram at
(37:45):
how to Lang for updates onfuture episodes and some other
fun stuff.
Thanks for listening.
Marina (37:52):
Bye.
Hello and welcome to How To Language a podcast about
language learning.
I'm Marina,
Jeremiah (00:07):
and I'm Jeremiah,
Marina (00:08):
We know you're probably sick of hearing about AI in the
news, but today we're divinginto the world of AI and
language learning how artificialintelligence is changing the way
we learn languages, what it getsright, what it absolutely
doesn't, and how to use it as atool without letting it replace
the most important parts oflearning a language.
Jeremiah (00:28):
Yeah.
And, don't worry, we didn't useAI to write this intro.
Mostly
Marina (00:35):
So we're talking about AI in language learning, but I
think before we dive too deep,we should explain what we mean
when we say ai.
'cause that is kind of a scarybuzzword that is thrown around a
lot.
But I think we should clarifywhat exactly we mean.
Jeremiah (00:53):
Yeah, for one thing, because when we say it, we mean
multiple things.
Kind of interchangeably, eventhough they're not
interchangeable.
And for two, I think, you know,people have been kind of
anticipating AI for probablysince 2001, a space oddity.
Odyssey, sorry.
I was getting my, my pop culturemixed up also.
(01:16):
It didn't come out in 2001.
That's just what it's called.
When did it come out?
1941.
I dunno.
I bring it up because wasn'tthere famously an AI in it?
Marina (01:27):
Yeah.
Hal,
Jeremiah (01:28):
how
Marina (01:28):
the ai Yeah.
Jeremiah (01:30):
I've never actually seen that movie.
Was Hal good or bad?
I
Marina (01:32):
think the whole point was that he was bad.
Yeah.
Jeremiah (01:34):
Usually the AI is bad.
Like if you've played PortalUhhuh, if you've seen Wally, if
you've seen W Yeah, that was a
Marina (01:42):
pretty direct reference.
Perfect
Jeremiah (01:43):
example.
If you've seen, um.
iBot.
The whole point of that moviewas that AI is bad.
But in all of those cases, theAI at least is, it's implied
that humans have figured out howto actually make an artificial.
Intelligence, it's in the name.
And one important thing to pointout is that in our timeline,
(02:05):
that is not what happened.
Like none of this is actually anartificial intelligence.
None of it is like a digitalbrain that does all the same
things your brain does.
We'll get into it, but like,it's fundamentally not what it
sounds like.
So I'll hand it back over to youto talk about what it actually
is mm-hmm.
At this stage in history anyway.
Marina (02:23):
So what we have currently, as far as I
understand it, is narrow ai,which means basically that it's
an artificial intelligence thatis designed and programmed to.
Do one specific task.
It doesn't have the greaterthinking capabilities that are
(02:45):
portrayed in movies like that.
Exactly.
Jeremiah (02:48):
Like you couldn't, put chat.
You couldn't download chat GPTinto like a robot body and then
have it discover fire orsomething.
Marina (02:56):
Right.
Like.
Chat t's only purpose that itwas designed for was to mimic
human language.
Jeremiah (03:04):
Exactly.
Yeah.
if you've ever played a videogame, all the other people in
the game that aren't real, allthe NPCs, they are all run by a
very narrow ai.
Like if you play Skyrim andyou're fighting bandits, that AI
has one goal, which is to killyou.
So you know, a perfect example.
It could be literally one onespecific task.
Yeah.
And it counts as AI apparently.
Marina (03:25):
Exactly.
And some of these AI tools aredesigned with language teaching
as their main task.
Some of them.
Can do that as a part of theirexisting tasks.
So, there's a few differentkinds of AI that we see in the
language learning landscape.
There's translation AI toolslike Google, Google Translate,
(03:49):
and Deep LI don't know if that'show you're supposed to say it.
If it's like de I don't thinkit's de, I dunno, there's so
many dumb names like thatnowadays.
There are a lot of dumb names.
You can never be sure.
So they like
Jeremiah (04:03):
chat.
PPT, sorry, I'm cutting thatout.
No, no, you can't cut that.
Okay.
Keep going.
Marina (04:11):
A lot of people don't think about Google translate as
ai, but it technically is andit's task is translation.
There's of course things likechatbots and natural language
processors.
So that's what people usuallyare talking about nowadays when
they talk about ai.
This is like chat GPT.
(04:31):
deep seek.
Grock, Elon Musk's ai.
Apparently.
There's writing evaluation toolslike Grammarly.
Those are super popular for.
Editing essays and giving youtips on writing and revision.
And then there's also thingslike what are called ITSs or
(04:52):
intelligent tutoring systems.
There's a lot of these fordifferent subjects, but the most
well-known one in languagelearning is Duolingo, which
isn't entirely ai, but it hasai.
Features in it that kind of helpassess your language learning
progress and adapt their lessonsbased on how you're doing.
Jeremiah (05:12):
Yeah.
It seems like what all thesekind of have in common, if you
think about it and part whatmakes them ai other than, oh,
they're programed to do a task,which is a lot of things that
don't count as AI is they all onsome level are like capable of
independently making decisions.
Even if it's really smalldecisions, like for Google
Translate, for example, it'sprogrammed to be able to decide.
(05:34):
How to word a sentence based oncriteria that are not obvious.
And for like Duolingo, I don'tknow, it decides like, when to
annoy you on your phone withyour push notifications.
Marina (05:47):
Mm-hmm.
Jeremiah (05:47):
And maybe we'll get to this, but speaking of chat bots,
they've started actuallyincorporating those in the
intelligent tutoring systemsnow.
Like memorize has little chat.
AI buddies that you can talk to.
I've tried the Chinese one andthe problem is it tries to, it
instantly uses words that Idon't know.
The problem with Chinese is youcan't just look at the words and
(06:08):
figure it out.
Unless you've seen the wordsbefore, it's like, might as well
be Greek for all, you know?
Marina (06:12):
So, Jeremiah, have you used ai?
It sounds like you have.
Jeremiah (06:18):
You know, I used to use Chate a lot and then I got
sick of paying for it when Irealized that Deep Seek can do
most of the same stuff for free.
So now I use deep seek if I havequestions.
And what else?
I had something else.
Speaking, yeah, experiences withai, actually when I was taking
old Norse in grad school.
(06:39):
For funsies.
I got kind of overwhelmed witheverything that I had going on
and I was like, I don't reallyactually care about this.
Like, I like going into class,but I don't, I'm not gonna like
stick with it enough to botheractually learning anything Oof.
Which is sad.
I really liked the instructor.
She was great.
But, so what I would do is Iwould just like feed Chacha Bee
(07:03):
my old Norse homework, mytranslation, what it came back
with was actually really good.
I got really good grades on it.
Marina (07:12):
That's exactly what you're not supposed to do.
I
Jeremiah (07:15):
know.
And here's my disclaimer forpeople listening.
If it's something you care aboutand you actually want to get
better at, don't use AI likethat.
You can still use it, but don'tjust have it do your work for
you.
That goes for anything.
But if you're taking one quarterof old Norse and you don't plan
on taking a second quarter orever using it.
You know?
Sure.
(07:35):
If you don't wanna fail, thenyou can use some ai.
I was surprised at how accurateit was though.
It was crazy even back then.
This was 2023, I think.
Marina (07:45):
Yeah.
Wow.
You just admitted to academicfraud.
Jeremiah (07:51):
I did what?
Marina (07:56):
All right, well, don't do that.
Jeremiah (07:58):
Don't do that.
Marina (08:01):
Okay.
So.
Ai, as a language learning toolis under this umbrella in second
language acquisition studiescalled Computer Assisted
Language Learning, which is, youknow, shortened to call,
(08:23):
everything's
Jeremiah (08:23):
an acronym these days.
I
Marina (08:24):
know, and it refers to.
The use of computers andtechnology to aid in the
language learning process,
Jeremiah (08:35):
which is so funny.
It's, it sounds like somethingthat, um.
Was like a research area in like2003 when computers were still
new and they were like, did youknow you could use computers to
learn a language?
And everybody was like, what?
That's crazy.
Marina (08:52):
I mean, yeah, that is kind of where it started.
Like, like CD ROMs that, oh mygod.
Like,
Jeremiah (08:57):
I don't even know what that is.
Marina (09:01):
So yeah.
But no, it really, it startedlike back then and then, you
know, it's.
Persisted all the way up untilnow with AI and things like
voice recognition software,things like that.
And so there's a pretty longhistory of research in Call that
has shown that technology can infact enhance language learning,
(09:24):
in several ways, which we'll getto shortly.
And that language learnersgenerally report positive
experiences and outcomes whenusing these tools.
So they are a good thing.
It's not like they hinderlanguage learning at all.
They tend to have the oppositeeffect.
Jeremiah (09:42):
Yeah.
It's not cheating.
Just like, you know, the firstfarmer who invented a plow
wasn't cheating.
Right?
Marina (09:48):
Yeah, exactly.
It's just another tool.
But you know, call hastraditionally been seen as a
supplementary tool, so it's nota replacement for real world
practice or language teachers ortextbooks or anything like that.
AI is just the latest evolutionof call with its own benefits
(10:12):
and limitations, which we'regoing to get into.
We'll talk about the benefits ofusing AI specifically in
language learning.
Things to look out for, thingsthat you're missing out on.
And we'll also go into some do'sand don'ts if you choose to use
AI as a supplementary tool.
So the benefits of using ai, Ithink the first one that comes
(10:36):
to mind and that's the the firstreason that a learner turns to
AI in the first place is thatit's an on demand tool.
It's almost like an on-demandlanguage partner.
Instead of going out and seekinga tutor.
Or visiting your teacher'soffice hours.
Um, you have immediate access toa partner to practice speaking,
(11:00):
writing and comprehensionwhenever you want, wherever you
are, which eliminates the needfor those human partners.
In theory it shouldn't replacehuman partners, but that is what
you can use it for in a pitch.
Similarly, you get immediateassessment on the output that
(11:22):
you generate, whether it'sspeaking to chat GPT or writing
your translation in Duolingo.
It gives you immediatecorrection and hopefully
explanation.
Addresses your mistakes in realtime, which is really a valuable
learning experience.
'cause making an error doesn'treally help you very much unless
(11:44):
you understand why it was anerror and how to correct the
mistake.
But in a classroom environment,you.
Can potentially receiveimmediate correction and
feedback from your teacher orfrom a tutor.
But that can also be scary.
It can be embarrassing.
It can be demoralizing to getthat from an authority figure,
(12:07):
somebody that you trust orrespect.
And so, you know, but AI is justan ai.
You can't really get yourfeelings hurt by it.
And so in that sense, it's avery low stakes language
partner.
You can talk to it or write toit and feel that you're not
going to be judged by them.
(12:28):
Many learners won't feel asanxious when speaking in their
target language.
It's very much a no judgmentzone, which definitely is a huge
benefit, if, especially if youstruggle with anxiety when it
comes to speaking your targetlanguage.
And I think the last mainbenefit of losing AI is that it
(12:49):
can be very good at adaptiveteaching, which means like it
can tailor its lessons or its.
Responses to be specific to yourneeds and your interests.
You can have a conversation withchat GPT about photography, if
(13:12):
that's your interest, in yourtarget language and be able to
talk about it at length, whereasyour teacher or tutor.
Might not be able to do that.
Um, they might not be able todevote that much time to it.
They might not know enough to beable to talk about it at length.
And so if that's something thatyou wanna do, that's something
that AI can provide.
(13:34):
It can also adapt to your level.
It can kind of get a feel forwhat level you're at based on
your output and adjust.
Its and adjust to, to that levelto make it more comprehensible.
Basically it can adapt to yourlevel and your learning styles.
(13:55):
And in fact, I think this isarguably the largest advantage
that AI has over traditionallanguage teachers and tools.
This is one of the reasons whysome teachers fear for the
future of the profession,because of the threat of ai.
I think that's not going tohappen, but it is something to
(14:18):
consider, because.
Teachers and other traditionalmethods like textbooks have to
cater to large swaths oflearners.
If you're teaching in aclassroom full of lots of
different people who are atdifferent levels and different,
learning styles, you can't tutoreach student individually.
(14:40):
So you have to kind of zoom outand do what's best for
everybody.
Now good teachers, part of yourjob is to do your best to
provide individual support andtake into account the individual
needs of your students.
But that can be really difficultdepending on the context that
you're in.
Jeremiah (14:58):
Mm-hmm.
A student or a,
Marina (15:00):
a teacher who has 10 students in her classroom,
that's a lot easier.
But if you have a multilevelclassroom full of like 40
students, like.
Good luck.
Yeah, that's really hard.
So, they can do their best, butit doesn't top an individualized
session with a, with a tutor,um, or in this case an AI tutor.
(15:26):
Because it can, you know,theoretically it can pick up on
what you're struggling with.
It can pick up on how you learnbest and then immediately apply
that to enhance the overallexperience.
The question though is whetheror not AI is capable of picking
up on those things and applyingthem effectively.
Jeremiah (15:48):
Yeah.
'cause you know that, that, thatis one of the consequences of
the fact that the AI is not likea functioning digital brain like
yours.
Like, like we have brains,right.
It's more like, a prettysophisticated, like pre
predictive text generator.
So the extent to which it cannotice things in the same way
(16:09):
that you and I can notice thingsis pretty, I would, I would say
dubious.
But on the other hand, you cantell it things like if you know,
oh, you know, this is what Iwanna work on, or whatever.
Ai, like chat bots in particularseem to do really well when you
give them a lot to work with.
Like, the more details youprovide, the more detailed their
(16:31):
response can be, becauseotherwise they're just working
with like the SumTotal of all ofthe training data that they
have.
But I think that segues uspretty nicely into, I won't say
the drawbacks of ai, and usingAI language tools, but more like
what you're missing out on, Iguess.
And I guess to start.
Maybe this is a drawback or moreof a consideration, I guess,
(16:53):
because some listeners mightjust have a really strong,
ethical, issue with even usingai.
In particular with usingchatbots, you know, whether
because they steal data orbecause they are hugely energy
intensive and those are fair.
So if you don't wanna use AIlanguage tools, I don't blame
(17:15):
you.
You're not gonna really missanything essential that you
need.
Whatever you're already doing isgonna be just fine.
However, if this is of interestto you, then by all means,
listen on.
So if you, if you are using AIlanguage tools, here's some
things you might be missing outon.
One thing I will say is thatsome AI tools like the Chatbots
(17:36):
have pretty advanced speakingand listening features, and in
addition to just writing, for awhile it was just writing, you
know, you would write to it andit would write back to you.
Kinda like, do you rememberClever Bot?
Marina (17:49):
No
Jeremiah (17:49):
way back in the day.
I don't know how that worked.
That was like a really,minimally intelligent, I guess
chat bot back in the day.
Back then that was just, thatblew our minds.
But you couldn't talk to it now,especially with chat GPT for
example, you can carry on a fullconversation with it, and it
almost feels exactly like you'retalking to another human.
(18:11):
And again, that's because it'snumber one job before anything
else is to convince you thatyou're talking to a person, not
explicitly, like if you ask it,are you a person, it'll say no.
But other than that, it's meantto create the illusion that
you're talking to a person.
So sometimes the AI model, youknow, the Shap PPT, will, even
(18:32):
include what we callparalinguistic features.
So those are things that are notwords, you know, they're not
part of the language itself, butthey're kind of things that go
around, the words that help givethe meaning, you know, the
expression that people say alot.
That 90% of communication isnonverbal, which is nonsense.
(18:54):
I mean, it's not really true.
But the kernel of truth to it isthat there's a lot of
communication that happensthrough body language and also
just like the things in speechin particular that are not
language.
So the pauses, like if you takean extra long pause between.
(19:15):
Your words to collect yourthoughts or, and, and, you know,
and these are things that, thatthe AI can do, right?
The, the chat, GPT, when youtalk to it, sometimes it'll
chuckle, which is kind ofcreepy.
That's creepy.
It'll pause.
Um, sometimes it'll say,
Marina (19:30):
Hmm,
Jeremiah (19:31):
or even clear its throat, which is really
unsettling.
However, it is important toremember that, ultimately it is
not another person.
So when you talk to a realperson, it's what it is, is it's
a relational process of give andtake where you are, you're
constantly listening andthinking and producing at the
(19:51):
same time in tandem with theother person who's doing the
exact same thing.
You know, it's like a weirdbalance where you ever find
yourself, like if you're in aconversation and you realize,
oh, I was just thinking aboutwhat I was gonna say next, and I
forgot to listen to what theywere saying, you know, that's a
constant, um.
It's a constant balance that youdo when you're in a conversation
with somebody.
(20:12):
And an AI does not have to dothat.
It's not thinking and it's noteven really creating what it's
doing is recombining contentfrom, like, the source is, it's
some total of training data,most of which it just kind of
stole.
So in other words, it'sessentially, like I said, a
really sophisticated predictivetext generator.
(20:34):
Exactly like the one on yourphone when you're typing and it
suggests the next word for you.
It provides a compellingillusion or like a facade of an
interaction, but it is just thatit's an illusion.
All of this is to say that whatyou're missing out on is the
real human interaction, which isone of the primary, like fuel
sources for language learning.
You can think of it kinda likethe difference between a pilot,
(20:56):
you know, when they're firstlearning how to fly and they go
in the simulator.
You ever see that?
They have like a flightsimulator.
It's really sophisticated.
Have I ever
Marina (21:03):
seen that?
Yeah.
If you
Jeremiah (21:04):
go to a museum, you know what, if we go to the
museum a flight, I bet we can.
I bet there's a flight simulatorthat you can use.
Marina (21:09):
That'd
Jeremiah (21:09):
be fun.
But yeah, they get in a flightsimulator, it's like a video
game, but it's reallysophisticated and it has all
those, all the right buttons andstuff.
Mm-hmm.
but that's what they do.
And you know, they do that totrain.
But if you were to just take apilot who has only flown in a
flight simulator and then putthem in charge of a commercial
(21:30):
airliner and told all thepassengers that they would freak
out, they'd be like, what do youmean?
This pilot has never flown areal plane.
So, you know, there's, what I'mgetting at is just like there's
a difference between learninghow to fly in a simulator versus
flying a real plane.
There's a really importantdifference between talking to a
simulated person and talking toa real person.
(21:50):
You can learn a lot from thesimulator, but the real
experience is gonna present youwith challenges, obstacles, and
problems that are emergent.
Which means they kind of justcome as if from nowhere.
They're not pre-programmed orwhatever, and unique.
Like problems that other peoplemight have faced, similar
problems, you might have facedsimilar problems, but ultimately
(22:10):
the conditions are unique toyour situation.
And so it's not so much that youlearn how to handle everything
that might come up.
It's more that you learn how tonegotiate obstacles kind of in
QuickTime as they come up.
And that is one of the mostimportant language skills that
doesn't really get talked aboutsuper explicitly, but we call it
(22:32):
negotiation for meaning.
I think we've talked about itbefore.
We'll probably have a wholeepisode about it someday.
That's an important skill andyou mostly can't learn it
without using the language inpractice, in real life
conversations.
Another gap in AI languagelearning is that real human
beings, when you talk to them.
They have specific backgrounds,they have memories, they have
(22:54):
cultural heritages, and theyhave finite linguistic resources
that are based on theirexperience in the lifetime that
they learned their language,right?
Like they don't know every word,or they might know a word, but
forget what it means.
Or they might have an idea, butnot sure what the word is for
it.
AI doesn't have that.
People don't always say whatthey mean.
(23:15):
Sometimes people say theopposite of what they mean.
That's where a lot of thoseparalinguistic features come up.
Like that's part of how you knowwhen someone's being sarcastic.
Sometimes they forget the wordthey're looking for or they have
difficulty explaining a concept,and all of that is stuff you
have to contend with when youtalk to someone.
People also have wants andneeds.
The person you're talking tomight be in a hurry or they
(23:37):
might be tired or they might befrustrated.
And all of these conditions arewhat makes every interaction in
your target language a uniquelearning experience.
And AI has no single culturalreservoir that it draws on.
It can't forget words or be in ahurry.
It doesn't have wants or needs.
So when you're talking to an ai,you'll never be presented with
(23:57):
the challenge of communicatingacross a cultural barrier or
having to be more concise orspeak more quickly, in order to
hold the attention of someonewho is in a rush or having to
ask clarifying questions to makesense of what someone's trying
to tell you when they aren'texplaining it very well.
Like these are all experiencesthat we've all had when talking
to people, even in our firstlanguages.
(24:19):
And AI really can't replicatethat.
Because that's not what it'smeant to do.
Lastly, some AI is good atmimicking paralinguistic
features, like I mentioned, solike a chuckle or taking a
breath.
It's kind of creepy, but itcan't replicate body language
because it doesn't have a body.
And so it's not very good.
It's also not very good atsubtext, which is another key
(24:41):
ingredient in most humancommunication.
So these are all some thingsthat AI just can't do.
And if you never go outside andyou never talk to someone in
your target language, just likewe always say, you have to go
outside, um, you'll be missingout on a lot of these things if
you just talk to Chad gt.
Marina (24:59):
So.
That was a lot of information,but, we'll kind of summarize it
here at the end with some do'sand don'ts for using AI for
language learning.
Definitely do ask it to breakdown difficult grammar points
into simple language or ask itto rephrase an unclear
explanation in like yourtextbook or something your
(25:21):
teacher said, this is one ofAI's strengths is that it's
really good at rephrasing andexplaining things in different
ways that might make more senseto you, especially if your home
base like a textbook is not veryhelpful.
You can
Jeremiah (25:37):
also ask it why a word or a structure in your target
language is the way it is.
As you know, like me, I likelooking up the etymology of
words in my target languagebecause I find that helpful for
connecting the dots betweenwords and kind of seeing the
matrix.
So if that's something that youfind helpful, AI can be helpful
with that.
However, you should not, assumeeverything that it says is true
(26:01):
if it's important information.
Something that you're gonna belike quizzed on later.
Uh, or especially if the AI'sanswer feels off, then
definitely double check what ittells you.
Marina (26:12):
I feel like AI would be particularly unreliable when it
comes to etymology.
I feel like it would be prettyprone to just making things up.
Jeremiah (26:21):
Yeah.
Or, or
Marina (26:22):
rely on folk etymologies,
Jeremiah (26:24):
potentially.
Yeah.
Because it gets all, most of itsdata from what the internet.
Marina (26:28):
I think so.
Yeah.
Jeremiah (26:29):
So theoretically.
In its data banks.
There's just a ton of Redditexplanations of things like, oh,
the, um, the etymology of the Fword is fornicating under
consent of the king.
I would not be surprised if, ifChad PT has confidently told
someone that
Marina (26:47):
Mm, yeah, it's, it's probably not going to the Oxford
English dictionary as its firstetymological source I'm sure it
Jeremiah (26:55):
has.
Reddit in its data, but youknow,
Marina (27:00):
yeah, I definitely don't take everything that it says as
fact.
Like we were talking aboutearlier, definitely do use it to
give you feedback on yourgrammar, your word choice, et
cetera.
It's gonna be really good atgiving personalized feedback
based on what you say or write,but don't rely on it to
translate everything, just likeyou wouldn't trust everything
(27:22):
Google Translates says.
There might be some translationtools that are.
Quote, unquote better thanothers.
Like deep L is gonna be betterthan Google, Google Translate,
for example.
But that doesn't mean that it'sinfallible, and it doesn't mean
that they don't have theirlimitations.
So, you know, I don't think thatwe'll be saying goodbye to human
translators anytime soon.
(27:44):
You always wanna double check,double check what it gives you
if you're using it for that.
Jeremiah (27:49):
Yeah, absolutely.
Also remember that, an importantthing with any language is that
it's a living thing unlessyou're studying a dead language.
It is a living unevenphenomenon.
And what I mean by that isthere's a whole corpus of
colloquialisms and idioms andslang terms that.
(28:16):
Are one temp temporallyconstrained.
And what I mean by that is likepeople weren't saying lock in or
crash out before, the last year,I think.
And you know, that's in English,but they're also often
geographically constrained,right?
Sometimes it'll be the sameword, but spelled differently in
different places.
(28:37):
Or mean different things indifferent places.
So what I'm getting at is,remember that AI does not
necessarily know everycolloquialism, every idiom, or
every slang term, especially ifit is particular to a certain
dialect or a geographical area.
Marina (28:56):
And if you do ask it to use.
Slang and colloquialisms from aparticular region of the world,
it might rely on stereotypes toachieve that, which isn't ideal
either.
Jeremiah (29:09):
Yeah, good point.
Like if you say, oh, speak Chileand Spanish.
Mm-hmm.
I feel like Chile and Spanish isun irrationally.
Look down on for some reason.
Marina (29:19):
Yeah, right.
Exactly.
So we don't wanna do thateither.
'cause it is just getting itsinformation from humans and
humans are fallible and havebias.
Mm-hmm.
And we don't want that.
So this last thing, in ourresearch for this episode, I
found, an article written bySchmidt and Strasser, Schmidt
(29:40):
and Rasser in, 2022 calledArtificial Intelligence in
Foreign Language Learning andTeaching a Call.
God for intelligent practice.
Jeremiah (29:53):
Oh no.
And call is like all capsbecause it's a, it's a pun.
Okay.
Anyway, off to a great start.
Marina (30:01):
At the very end of this article, they, give this utopian
their words vision of what a,computer assisted foreign
language classroom might looklike in 2040.
And so we have an excerpt here.
I shortened it'cause it waspretty long, so.
Mm-hmm.
I just, just keep that in mind.
(30:22):
It's not a word for word.
But
Jeremiah (30:26):
yeah.
And it also just glancing overit, it looks like there's some,
academicy jargon in here.
Mm-hmm.
Not, not crazy, but a littlebit.
Yeah.
How about I, I'll just go aheadand read it out loud because
this looks pretty interesting.
Cool.
I love eviscerating academicjargony, writing because I
always find it so.
(30:47):
You silly.
All right, here we go.
It says in 2040, the foreignlanguage classroom will be an
evidence-based, Hey, that's us.
We're evidence-based and we'rejust based, evidence-based
blended learning environmentwith bring your own device
solutions in all classrooms andsubjects.
(31:10):
Is everybody gonna be able toafford those devices?
I don't know.
Anyway, in 2040, the printedtextbook will have become an
interactive, multimedia,adaptive learning and practice
environment that is perfectlyadapted to the needs of both
face-to-face teaching and thephases of individual practice
and self-directed learning.
(31:31):
Okay.
The digitally enhanced classroomof the year 2040 will help to
diagnose needs and learningprogress, providing direct
access to differentiated andneeds-based support services.
The classroom of the future willcleverly combine the advantages
of digital learning with provencomputer free methods.
(31:51):
Content and tasks forface-to-face teaching, which
will remain indispensable andhighly significant for
successful learning.
And even in 2040, we will stillhave and need teachers,
well-trained data literateteachers who are competent,
critical, and reflective intheir use of media and
technology support, and who willuse and who use empirically
(32:13):
established digital scenarios,uh, that are based on valuable.
Learning content and are linkedto meaningfully constructed
challenging learning tasks andexercise opportunities designed
to support individual learning.
Oh boy.
Okay.
What, what are your thoughts?
Right off, right off the bat.
Marina (32:32):
That last one was one sentence.
Oh my.
Jeremiah (32:34):
That whole thing.
Wow.
Okay.
Marina (32:37):
Um, it is very jargony.
But you know, in the context ofthe rest of the article, it
makes callbacks.
And makes more sense,
Jeremiah (32:44):
makes callbacks.
Oh God.
Marina (32:49):
Um, I think it's basically saying that the goal
here is an adaptive learningenvironment that still makes
good use of quote unquotecomputer free methods.
And still necessitates teachers.
(33:10):
So, you know, I think that'swhat they mean when they say an
evidence-based blended learningenvironment, for all classrooms
and subjects.
And so I think that that'sachievable for sure.
I think we're already kind of onour way there.
You know, it'll.
Be nice to have, interactive,multimedia, adaptive learning
(33:32):
for everybody.
That might be kinda tough by2040.
Um, Do you have any thoughts?
Jeremiah (33:41):
Yeah, I mean, I'm just thinking about do you remember.
In, I don't know if Japanese hadthis, but Spanish had this, when
I was taking Spanish inundergrad, whatever textbook we
had had its own like proprietaryhomework website Yeah.
That you needed a code to getinto, and that's where you
(34:02):
submitted your homework and likea lot of the class was just on
there.
From what I saw that did notchange one iota in my entire
college career.
I don't think it's any differentnow.
I feel like the textbookcompanies still have a monopoly
on, that, homework site.
When I think of actuallyexisting digital.
(34:25):
Whatever in the languageclassroom.
That's kind of what I think of.
Marina (34:30):
But that's not necessarily AI either.
No, but I mean, that's just adigitally enhanced classroom.
Jeremiah (34:36):
That's true.
I'm just, I'm just basing thisoff of what this thing says.
cause this doesn't talk about AIthat much, this little excerpt.
Marina (34:44):
No, it really doesn't.
In fact.
Jeremiah (34:48):
Which I think kind of gets to a point that I'll make,
which is that actually that youmade a while ago, which is that
this whole AI craze is just likeone more step in a long standing
trend in language learning andin learning in general to
incorporate, you know, quoteunquote technology, which is
(35:10):
funny because it's always been.
It's always incorporatedtechnology.
I think they just mean digitaltechnology.
So in that sense, AI isn't supergroundbreaking.
It's just like kind ofqualitatively differentiated by
how weird and futuristic it is.
Marina (35:27):
Yeah.
The full, the full section ofthe article does mention AI
looks like twice
Jeremiah (35:33):
Mm.
Marina (35:34):
For example, quote, the learning platform deployed in
2040 is thus an AI based digitallearning support system.
A resource, a tool for studentsand teachers, which will be used
to add value to learn to thelearning process.
Hmm.
And then later they say, quote,a sustainable approach to make
teachers become acquainted withAI technology within a rather
non technocratic narrative,would probably, et cetera, et
(35:57):
cetera.
but they do mention ai, butreally not that much for an
article that's entirely aboutai.
Jeremiah (36:03):
Yeah.
Marina (36:03):
But I think, yeah, I think that that does speak to
the fact that AI is just onesmall component of a digitally
enhanced or call approach tolanguage learning.
You know, it's sort of thebuzzword now, but you know.
It really is just one component.
Jeremiah (36:21):
Yeah, I guess I really don't have that many gripes with
this article.
Then other than that, they'rekind of just yapping about, you
know, basically what we have nowexcept without, weird.
Monopolist, textbook companiesand technical difficulties,
which I feel like is one of themain issues with any digital
(36:44):
anything in the classroom, isthat it's always a pain to use.
Like canvas.
Oh my god.
Mm-hmm.
Um,
Marina (36:52):
yeah, I think, different platforms and tools have
different strengths and I don'tthink that there's really
anything inherently wrong with.
Combining them?
No,
Jeremiah (37:02):
no.
Marina (37:02):
You know, instead of just relying on the digital
textbook that gives you shortfilms to watch and makes you
order these words in the rightorder to make a sentence for
your homework kind of thing.
I think that if we could makeuse of all the best parts of
technology and the best parts ofAI for the particular learner,
Jeremiah (37:24):
that would be ideal.
Well, thank you for tuning intothis episode of How to Language.
We hope that you're walking awayfeeling better equipped to use
AI thoughtfully in your languagelearning if you want.
And if you like what we do, youcan rate and review us on your
preferred podcast app and youcan follow us on Instagram at
(37:45):
how to Lang for updates onfuture episodes and some other
fun stuff.
Thanks for listening.
Marina (37:52):
Bye.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!