AI and language learning - What have you tried?

[CaroleR]

Ah, but do you maybe remember it simply because the AI failed to give a good example, so you spent a lot of time engaging with its meaning while thinking "no, that's not right... no, that isn't either"...? Another one ofmy bugbears with a lot of language learning advice (which I've moaned about with mnemonics in particular) is that people often give examples of the technique, and then when I point out that the example itself is pretty weak, they respond "ah, but it's just an example". To that I usually respond that if you think a technique is good, you should be able to come up with a good example to demonstrate it. If you can't come up with a good example, are you really teaching people to use the technique well...?

That could very well be the reason I remember the meaning of the word. But I was simply relating my one-time experience of what I had tried, as requested by the OP. I said it could be useful and suggest that it's something people may want to try. Or not. Make of it what you will.

Here's the image. Sorry, couldn't figure out how to place it where I wanted it to go.

That'll be because you used the quote feature to format the ChatGPT exchange. The forum software (phpBB) interprets quote tags as being quotations from an earlier post in the thread, and I believe that it strips img elements out of quoted text on the grounds that there's no point showing the same images loads of times when you get quotes of quotes...

Thanks! I thought that might be the reason but wasn't willing to spend any more time trying to figure it out.

[Cainntear]

I was not aggressive with Le Baron, I was sincerely baffled that he could think that there was no hard work in my approach while I explicitly stated that I use AI to generate lists to drill verb forms.

Are you trolling me right now? Or do you seriously not understand that this sort of sentence generally comes across as highly aggressive?

I then explained my point via the "exercises with answer keys" examples.

But what you're doing here is blaming us for not understanding, and telling us "well I said XYZ, so you should have understood me." Arguing about what did say doesn't address tha matter of what you didn't. If your explanation is not complete, it leaves the reader to presume things to fill up the gaps with. We filled the gaps with incorrect presumptions and you are unhappy. If we had by happy coincidence filled in the gaps with correct assumptions you would have been happy. Why? That's like choosing emotions by rolling dice, because it's all pure chance.

My "aggressiveness" with Le Baron in reality is my passionate desire to defend effective tools which facilitate language learning and which, quite literally, allowed me to finally progress with Latin (which up to a recent point had been gatekept by Latinists, at least in my country).

Well you did a poor job of defending them. Explaininjg the misunderstandings is more effective than shouting at people for failing to presume the right unspoken assumption.

You, on the other hand, really pissed me, I don't deny that, because you clearly didn't bother to read my posts and started to talk to an imaginary person. Like, for instance, now re grammar

OK, but... did you say that?
<...>
In your case, you didn't mention reading grammar books explicitly.

It's literally in my original message!

Sure, you need to have a firm grasp of grammar

That is literally different, and again you're angry that someone dares to fail to understand your message rather than looking for why they fail to understand.

The sentence you have quoted is in the middle of your post, and you treat it as an aside: an unimportant throwaway comment. But it's not unimportant -- absolutely the opposite. It is vital that everyone knows where to start from. What does "Sure, you need to have a firm grasp of grammar" actually mean? There's a lot unsaid there.

You know, you also could've chosen a different tone in your response and my reaction could've been different. Like, asking clarification questions, even aggressively. I don't mind aggressiveness btw, as long people attack actual statements, not the imaginary ones.

Are you trolling me right now? Or do you seriously not understand that my tone was led by your tone?

[JLS]

Well, I'm happy that we all take our language learning seriously. I'm happy too that we all desire to do the best we can in it.

Still, the arguing isn't necessary. Misinterpretations happen. Mistakes happen. Better to just let it go.

Again, positive uses of AI in language learning, along with whatever guardrails you may choose to employ to prevent error.

[Cainntear]

Ok, so I think the key thing that hasn't been said particularly directly in einzelne's post, but is a very strong undercurrent, is that AI can allow us to do things that we want to do, but that aren't practical given the time.

When we look at grammar books, we're training declarative memory and can record the rules, but we don't necessarily train our procedural memory -- the bit that let's us do things automatically. His objection to Latin teaching was that it's often *only* declarative and not procedural, and when he talks against that, it sounds as though he's taking against declarative work,when he's actually taking about filing in the gaps by doing procedural work *after* declarative work.

There are lots of things that I *could* do by producing multiple learning examples of my own by applying the rules and then using the examples for repeated practice, but at the end of the day, that's long and slow, and it would take me a very long time and just wouldn't be worth it.

But if I use an AI to generate examples, I absolutely can.


The thing that I think einzelne was trying to say was that if you are getting an AI to generate examples, they have to be examples that you could have created yourself by consciously applying declaratively learned rules, but that you couldn't spontaneously create from procedural language knowledge. When you do that, you're able to verify that it is good enough, and it is quicker to use declarative knowledge to verify something than to write it from nothing. This means that it may be practical now to create your own practice sets with the help of AI as a single stage in your learning, but you can't expect AI to teach you the whole thing.

[JLS]

Ok, so I think the key thing that hasn't been said particularly directly in einzelne's post, but is a very strong undercurrent, is that AI can allow us to do things that we want to do, but that aren't practical given the time.

When we look at grammar books, we're training declarative memory and can record the rules, but we don't necessarily train our procedural memory -- the bit that let's us do things automatically. His objection to Latin teaching was that it's often *only* declarative and not procedural, and when he talks against that, it sounds as though he's taking against declarative work,when he's actually taking about filing in the gaps by doing procedural work *after* declarative work.

There are lots of things that I *could* do by producing multiple learning examples of my own by applying the rules and then using the examples for repeated practice, but at the end of the day, that's long and slow, and it would take me a very long time and just wouldn't be worth it.

But if I use an AI to generate examples, I absolutely can.


The thing that I think einzelne was trying to say was that if you are getting an AI to generate examples, they have to be examples that you could have created yourself by consciously applying declaratively learned rules, but that you couldn't spontaneously create from procedural language knowledge. When you do that, you're able to verify that it is good enough, and it is quicker to use declarative knowledge to verify something than to write it from nothing. This means that it may be practical now to create your own practice sets with the help of AI as a single stage in your learning, but you can't expect AI to teach you the whole thing.

That's where I find AI useful. Great for generating exercises. It'd take too much time to write my own. And by the time I'm done writing them, I know them.

[Le Baron]

Well frankly I don't think it takes all that much time and that when I am doing that I'm involved in the language. The fixation with trying to be 'efficient' through AI is ridiculous.

[einzelne]

The fixation with trying to be 'efficient' through AI is ridiculous.

I don’t know why you keep ignoring the fact that I have to work on my Latin under severe time constraints.

— Hey! I got a dishwasher and a smart vacuum cleaner. Now I have an extra hour everyday, so I can finally train for a half-marathon! I’ve been dreaming to run it for years but I could only train and participate in 5k and 10k runs...
— Ugh, I don’t know why are you so fixated on smart technologies… it's ridiculous.

Ok, one more time to make my point clear. Step by step.

You read a sentence in the present tense and want to drill it in other tenses as well. Without AI.

1. You open your browser. — do you engage with the language at this point?
2. You type wiktionary.org — do you engage with the language at this point?
3. You type the verb. — do you engage with the language at this point?
4. You click on the link. — do you engage with the language at this point?
5. You a a page with this word in different languages, sometimes it can be a dozen. — do you engage with the language at this point?
6. You scroll to find the entry about the Latin word — do you engage with the language at this point?
7. You click on the table to open it. — do you engage with the language at this point?
8. You start copying one verb form after another. — ok, here you finally start to engage with the language. but it’s just passive recognition at this point
9. Since words are interlinked, sometimes you accidentally click on the word. You curse, click the back button. — do you engage with the language at this point?

The whole process can take 40-60 seconds.With the AI generated list at hand, you could practice active recollection all this time. So I could get a 15-20 min intense section, non stop. You literally feel like these forms are branded into your brain, because you’re are not distracted but dumb mechanical work. (You like to write these forms by hand? Fine, you can do the same with with these lists. I just think it's not the smartest time management.)

Alexander Arguelles says that the harder the language, the higher the level you want to get, the more time you need to practice it. I know it, you know it, it’s an open secret. AI helped me practice with enough level of intensity to get to another level. Something which wasn’t possible for me before.

Literally, I keep telling you: look, my coach told me that to get to another level, I need to go to the gym and lift weights for 1 hour, but it takes me 15 minutes to get to the gym, so I can only get 30 min of training. Now, finally I moved to an apartment complex with a gym (AI) and I train for 1 hour and can see the results!
You response: duh… why are you so fixated on than???

Finally, I can only quote Geoffrey Steadman’s preface to his readers where he argues against the stupid tradition of not providing translations and vocabulary lists for Latin texts. Looking up a word in a dictionary is not a big deal, right? Som what’s that all the fuss about?

One of the virtues of this commentary is that it eliminates time-consuming dictionary work. While there are many occasions where a dictionary is absolutely necessary for developing a nuanced reading of the Greek, in most instances any advantage that may come from looking up a word and exploring alternative meanings is outweighed by the time and effort spent in the process. Many continue to defend this practice, but I am convinced that such work has little pedagogical value for intermediate and advanced students and that the time saved by avoiding such drudgery can be better spent reading more Greek, reviewing morphology, memorizing vocabulary, mastering verb stems, and reading advanced-level commentaries and secondary literature.

That's it. That's what AI did to me. Eliminated drudgery (by providing parallel texts and generating fine-tuned drills) and increased the time of meaningful engagement with the language.

Feel free to continue to ridicule my "fixation" while I'm finishing drilling the Subjunctive forms and slowly preparing to read untranslated works of Leibniz in the original this fall.

[Querneus]

As someone who's sometimes been involved in editing Wiktionary, this whole discussion is funny to me because of the inconveniences of Wiktionary. Einzelne's disappointment with Wiktionary's UI is legit, and for a long time there's been voiced complaints about how you can't just filter out the languages you care about (no Asturian or Catalan needed if all you need is scrolling to Portuguese), how you can't limit the search to certain languages, how you can't just have a setting where tables are displayed by default... Desiderata that maybe one day we'll have fulfilled but it's low priority for the Wikimedia Foundation devs.

And it matters for Latin because English Wiktionary is one of the very few resources out there at all that will actually give you conjugated verbs (and declined nouns and adjectives) to a pretty reliable extent. With vowel lengths indicated, to boot. Dictionaries typically don't even give you the principal parts in full form (e.g. "ungo (3) -nxi -nctum" for what should be "ungō ungere ūnxī ūnctum").

No wonder Einzelne finds it easier to just ask a large language model "AI".

[Le Baron]

I don’t know why you keep ignoring the fact that I have to work on my Latin under severe time constraints.

Well I'm not, but I wasn't even referring to you in particular. Nor that kind of 'efficiency' in particular. I read your explanation before and said I understood it, even if I don't agree with all of it.

And what's that? 'Severe' time constraints you say? Are you sure you're learning anything under such harsh and exacting conditions? Thank heavens for AI.

I have a home gym (and no 'coach').

[Cainntear]

You read a sentence in the present tense and want to drill it in other tenses as well. Without AI.

1. You open your browser. — do you engage with the language at this point?
2. You type wiktionary.org — do you engage with the language at this point?
3. You type the verb. — do you engage with the language at this point?
4. You click on the link. — do you engage with the language at this point?
5. You a a page with this word in different languages, sometimes it can be a dozen. — do you engage with the language at this point?
6. You scroll to find the entry about the Latin word — do you engage with the language at this point?
7. You click on the table to open it. — do you engage with the language at this point?
8. You start copying one verb form after another. — ok, here you finally start to engage with the language. but it’s just passive recognition at this point
9. Since words are interlinked, sometimes you accidentally click on the word. You curse, click the back button. — do you engage with the language at this point?

The whole process can take 40-60 seconds.With the AI generated list at hand, you could practice active recollection all this time. So I could get a 15-20 min intense section, non stop. You literally feel like these forms are branded into your brain, because you’re are not distracted but dumb mechanical work. (You like to write these forms by hand? Fine, you can do the same with with these lists. I just think it's not the smartest time management.)

Your whole tacit assumption is that the answer to "do you engage with the language at this point?" each time is no, and the deeper tacit assumption is that if you're not engaging with the language, you're not doing anything useful.

This is what I mean about focusing on the superficial activity, and you're ignoring the point that the brain can be active in different ways. At the very least, the process of working to find something does force you to focus on the thing you're looking for, and the act of focusing on it at the very least gives your brain justification to recognise that it's important. But more than that, most of us really have no idea of what our own brain is doing in the background. If the person isn't bored and doesn't experience the process as "mechnical work"... maybe it isn't. Maybe their brain is doing something, even if they don't know what that something is.

In educational research, there's such a thing as a "delayed post-test", a test done a week or two after the teaching, and crucially with no additional practice done. Very often, participants do better in a delayed post-test than an immediate post-test (i.e. a test taken at the end of the training session or one day after it). In fact, this happens so often that if participants don't do better in the delayed post-test, it's a pretty strong indicator that their performance on the immediate post-test was most likely just rote memorisation and not any deeper learning than that -- deeper learning continues as background processing of the information to be learned.

So yeah... we learn over time regardless of what we're doing, and if we try to find ways to improve the apparent efficiency of the superficial tasks by getting more superficial tasks completed in a designated time slot, we might indeed be making it less efficient by robbing the brain of the time and attentional resources it needs to perform the background processing.

This is kind of like cramming vs learning. If you cram in the run-up to an exam, you can regurgitate the facts in the exam hall a few days later, but the brain isn't given time or space to form a model of how the facts act together as a system.

Alexander Arguelles says that the harder the language, the higher the level you want to get, the more time you need to practice it. I know it, you know it, it’s an open secret. AI helped me practice with enough level of intensity to get to another level. Something which wasn’t possible for me before.

Why should we defer to Arguelles's opinions?

Literally, I keep telling you: look, my coach told me that to get to another level, I need to go to the gym and lift weights for 1 hour, but it takes me 15 minutes to get to the gym, so I can only get 30 min of training. Now, finally I moved to an apartment complex with a gym (AI) and I train for 1 hour and can see the results!
You response: duh… why are you so fixated on than???

OK, so let me kill your analogy. My personal trainer gives me a 1 hour programme, but due to travel time to the gym and other commitments, I can only spend 45 minutes in the gym. The 1 hour programme my PT provided has approximately 15 minutes total of rest time in between activities. Does that mean I can cut out the recovery time between sets and do everything more efficiently in 45 minutes? No -- the recuperation is in there for a reason: recovery is necessary for your muscles to perform well and to repair and adapt.

If you think that this is a false analogy for brain activity... well that undermines your analogy in the first place.

Finally, I can only quote Geoffrey Steadman’s preface to his readers where he argues against the stupid tradition of not providing translations and vocabulary lists for Latin texts. Looking up a word in a dictionary is not a big deal, right? Som what’s that all the fuss about?

One of the virtues of this commentary is that it eliminates time-consuming dictionary work. While there are many occasions where a dictionary is absolutely necessary for developing a nuanced reading of the Greek, in most instances any advantage that may come from looking up a word and exploring alternative meanings is outweighed by the time and effort spent in the process. Many continue to defend this practice, but I am convinced that such work has little pedagogical value for intermediate and advanced students and that the time saved by avoiding such drudgery can be better spent reading more Greek, reviewing morphology, memorizing vocabulary, mastering verb stems, and reading advanced-level commentaries and secondary literature.

That's it. That's what AI did to me. Eliminated drudgery (by providing parallel texts and generating fine-tuned drills) and increased the time of meaningful engagement with the language.

Right, so let's talk about dictionaries a bit more, because they're a good example of where I'm coming from. In the early days, dictionaries were on paper, and looking them up was a nuisance and a bit time consuming. Then electronic dictionaries came out. Why do I still have a pile of paper dictionaries...? Because when I started using electronic dictionaries, I came to realise that the words I looked up in paper dictionaries were easier to remember than the words I looked up in electronic dictionaries. It seemed to me that the act of looking up a word in a paper dictionary was enough of an inconvenience to motivate me to remember it, but when I could just click on a word and see its translation, I had no motivation to remember. In fact, to answer your question of do you engage with the language at this point?... no. I found myself switching off and putting no effort into learning the word. Hell, my eyes would just gloss over it without even trying to process the pronunciation, because the easiest way to understand was just to click on it.

There have been browser plugins to do immediate dictionary lookups for years now. I haven't used them. I did finally install a translation plugin about a year ago that offers options of full page translations or on-demand translation of words and phrases, but I almost never use it, and never for actual language learning.

Feel free to continue to ridicule my "fixation" while I'm finishing drilling the Subjunctive forms and slowly preparing to read untranslated works of Leibniz in the original this fall.

It's coming across as a fixation because you're talking about very superficial descriptions of activities and presenting them in very absolutist terms, as though these activities are the "one true path". As I keep saying, your brain is doing more that you know, and also, you're doing more than your telling us. You say you're telling us what you do, but it looks like recommendations or advice for others and you haven't given a complete enough story for us to replicate your advice.

I have attempted to interpret and give a clearer description of what I think you're getting at, but you have neither confirmed nor denied that.