Best Combo of Anki and LingQ: ChatGPT

So if you just read random material you will not get enough repetitions to get past the intermediate gap without reading millions upon millions of words.
The old way around this was to read graded readers with a focus on specific vocabulary levels.

An alternative old way was to memorize isolated frequency words with anki.

There is an argument to be made (I believe this myself) that an ideal combination is to do anki plus lingQ because you’re getting enough repetitions specifically targeted to those you need more work on plus you get lingQ exposure to language in it’s whole form.

Unfortunately the two are not strongly correlated: you don’t get lingQ focus on the words that you’re having a hard time with in anki. Some suggest using sentences. I personally disagree with sentences for a couple reasons but I won’t get into them. Regardless, even the sentences won’t correlate with lingQ.

Others have suggested (likely correctly) that a stronger memory trace would be formed if you were able to learn the words you are targeting specifically if you encounter them in their natural form. i.e. in written or spoken material.

So what occurred to me a couple days ago is to specifically get chatGPT to write you some short stories with the words you are practising each day. So you therefore get not only the anki practice but you also get to see the words you’re currently working on, in their native form.

So anyhow, I’m now doing that in addition to reading on lingQ. It’s costing me about half an hour extra a day to generate some stories containing my today words and read them in lingQ.

Double whammy. If I’m right about this, I expect to level up solidly over the remaining three months of my current six month challenge.

Stay tuned…

EDIT: For pedants like my friend Peter (and I say that in the best possible way, not as a dig - Peter will keep you straight and coherent) - I’m talking about comprehension ONLY, not output. So intermediate → advanced comprehension (reading/listening) shortcuts.

7 Likes

I don’t disagree with your thoughts here, It’s inefficient to read random materials as you don’t get the exposure where you need it. However, do you really need 100% comprehension of words you already have been exposed to? You may not be mastering every word with random text, but you are still being exposed to new words and many of those are sticking, Put another way, you have less exposure with random materials, but if those words are not appearing in random materials, maybe you’re wasting your time trying t memorize them with ANKI.

I have another theory about this and a different approach. I’ve found that sticking with a particular author or a particular genre increases the repetition of words. Most authors keep the same style, expressions and word choices in all their writing. So I see the same words and phrases coming up all the time. Same within a genre. Each has certain elements that fans of the genre expect, so the same words, situations, etc. keep appearing.

There just are no shortcuts. You learn when your brain is ready to learn. Keep it fun, interesting, intense enough and regular and just about any method words in the end.

7 Likes

“There just are no shortcuts.”
Thanks for your opinion. But there actually are. I already know what the shortcut is to get to intermediate. I haven’t (yet) found the shortcut to get to advanced. I’m betting, however, if there is, that this is it.

EDIT: You are in fact, correct about sticking to the same authors/same genre though. Reading everything in Harry Potter or some other series will give you the words you need for that series. Also… if you think about it a bit more, your method is in fact a shortcut.
My belief is that shortcuts to get to intermediate are more abundant than shortcuts to go from intermediate to advanced. The former can be done in a variety of ways (Benny Lewis, Steve Kaufmann, Luca Lampariello, IKenna etc are some examples). As far as I know, though, the only “shortcut” up till the creation of AI to go from intermediate to advanced is the English guy who learned Mandarin in a single year. And his method won’t work for me.

2 Likes

Yes, I follow the same strategy. For example, when I am reading a physical book I write down all the unknown words in a notebook. I create short stories, minimally, I use 10 new words per story. If some words repeat a lot during the course of reading the book (whether they are the favorites of the author or important ones in explaining thoughts). I also create dialogues plus stories with these important words and import them into LingQ. This way I will get more repetition in context and keep the content interesting. I have gradually moved away from Anki, though.

It was a great feeling for me to learn a new word that ChatGPT used. I came across it in a short story the first time around, then I came across it again while reading a physical book. In a real conversation with a German native, my subconscious mind came up with it and the German native also used it to express her thoughts. I really liked this chain reaction that was started by ChatGPT.

I can dedicate only limited time to this extra activity so I have to choose between adding sentences to Anki or creating stories by ChatGPT. I opted for a latter option.

3 Likes

“I already know what the shortcut is to get to intermediate”
Hm, I’m not so sure. You seem to have struggled for more than 1 year to understand
how the case system in Russian works.

No offence, but that’s not a “shortcut” in my SLA book (neither from a learner nor from a teacher perspective). That’s rather the wrong “mix” of approaches…

1 Like

Lol Peter’s classic bluntness.

You’re making a straw man argument bro: You don’t need the case system to understand at an intermediate level. You only need the case system to be able to output at an intermediate level. Plus I’ve been doing way less hours than my initial burst and speculating that the number of hours isn’t enough. Also nowhere have I said I’m intermediate at speaking, which has never really been my goal, so I haven’t really placed a great deal of effort into trying to speak. Maybe a handful of hours at the most.

I also know fine well (and so should you if you’ve been paying attention) that if you immerse and force yourself to speak with natives for at least 5 hours a week you will get to advanced.

So back to the actual point: That’s not the quest here (and perhaps I haven’t framed it correctly). The quest here is how to get to advanced comprehension as a shortcut. LingQ just by reading novels isn’t it (as a shortcut).

I have added an edit for others of your ilk my friend.

“Lol Peter’s classic bluntness”
Sure, xxdb: I mean I’m German and at the same time an IT guy. Both are known for not beating around the bush.
And then there’s the “Peter genetics part”… :slight_smile:

BTW, this “mix” works extremely well for troubleshooting and problem-solving…

“You don’t need the case system to understand at an intermediate level. You only need the case system to be able to output at an intermediate level.”
My (wrong?) impression (see also your comment on how often you should repeat the “Mini Stories”) is that you tend to avoid grammar altogether and just hope that you’ll understand the grammar structures with enough exposure to your L2.

Yes, this works for native speakers because - from a learner’s point of view - their L1 exposure time is insanely long. However, this is an inefficient strategy for language learners.

Grammar-oriented approaches can be a “shortcut”, esp. for distant L2s, in this SLA context - provided they are “light” approaches…

That’s all.

1 Like

“The quest here is how to get to advanced comprehension as a shortcut.”
Sure.

The shortcut is, in my opinion a bi- or multi-modal approach à la fast-paced “reading while listening”
(see: Multi-modal language input: A learned superadditive effect) based on a collection of ca. 1000-2000 texts full of authentic everyday dialogues (YT, podcasts, Netflix, etc.) in the L2.

And the focus on an IM level should be “collocations”, but definitely not “single words”…

The ideal SW-based SLA solution would be based on such L2 collections and collocation-specific (gamified?) training (which could involve generative AIs - but, probably, less for generating stories and more for using these collocations in interactions)…

2 Likes

@Peter:
Grrrr. Your other comment I couldn’t find anything to argue with. That sucks.

And the focus on an IM level should be “collocations”, but definitely not “single words”…
Of course. LingQ should give you that. And it is. Progress at intermediate isn’t zero. It just feels slow. And that might be expected because it’s diminishing returns. So in essence I’m not really doing 1-2 hours. Comparatively speaking I’m actually doing 1/2-1 hour because each hour only covers half the volume of new words than below intermediate.

The ideal SW-based SLA solution would be based on such L2 collections and collocation-specific
Maybe. That said, something I have noticed as an analogue to this suggestion is the chatGPT generated stories around the words. The words come with very specific collocations. As you would expect… So essentially by generating stories for consumption in lingQ (and the corresponding audio), you are getting the collocation exposure for those precise words that you don’t know. Further locking it in.

Anyhow, like I’ve said a couple times now, the proof will be in the stats. If my retention stats improve then I can definitely say my hypothesis worked. If they remain the same or get worse then the hypothesis failed. Since it is a single change and I’m doing everything else the same it is a 100% valid scientific experiment.
On a sample of one.
Ha.

1 Like

A riff on your “all the unknown words in a book”…

Before I read the book what I do is this:

I export all the words into a frequency list, find those with at least two repetitions. I then remove the names (e.g. countries, cities, first names, last names).
My reasoning is that those with at least 2 repetitions are more likely to be useful than those with a single instance. Single instance words could be words that you’ll never see again in your lifetime. No way to tell.

Anyhow. Next step is I take this list back into LingQ, have lingQ identify just those words which I don’t know already. That’s usually down to less than a hundred, more often than not it’s about 50 words. Then I skim read to see if they are versions of a word I already know but with a different declension or conjugation. I remove those. Then I’m left with the honest-to-goodness words I have never seen before that are repeated at least two times.

Then… I add those to my anki stack before I start to read the book.

1 Like

“Progress at intermediate isn’t zero. It just feels slow.”
I’d say all this IM grinding is both an “expectation game” and a matter of the “distance / closeness of the L2”.

Besides, I think progress at an IM level isn’t so slow if we use all the L2 levers at our disposal:

  1. Using a collection of > 1000 everyday dialogues (in my case: in English/Spanish), which I wanted to translate in bulk into other L2s I’m studying (in my case: Br Port, Dutch, and maybe Japanese). I’ve been collecting such dialogues since Aug. 2022.

  2. Using my go to SLA books that I know inside and out.
    a) non fiction: the Harari trilogy
    b) fiction: beststellers such King’s “It” (with ca. 500k words), etc.

  3. Using an ultra-listening-while-reading approach (speed ca. 1.5x or higher) with 2 Pomodoro blocks à 25 min a day of focused attention at an IM level

  4. Using LingQ-to-Anki for testing my retention from 3) or interacting with generative AIs, ca. 5-10 min.

  5. Optional: free listening when running in the morning or watching Netflix, etc. = ca. 0.5 h

So, the “Minimum Effective SLA Dose” would be:
Goals:

  • 2 Pomodoro blocks = 50 min with ca. 10k words read and listen to
  • Overall words digested for closer languages from the Romance or Germanic language families: ca. 2 millions (in addition to the words already acquired on A1-A2/B1 levels).

→ 200 days à 50 min + 10 min with 5) + 6) → 200 h to reach a B2-C1 level in listening + reading comprehension (that’s my hypothesis for my Dutch challenge 2023 at the moment) + 0.5 h * 200 = 100 h of free listening (point 6) → altogether: ca. 300 h for Dutch = ca. 1 h and 1.5 h per day, respectively as a “Minimum Effective SLA dose”.

My thesis:
This could work for all Romance and Germanic languages (based on my native tongue German and the L2s I already know).

My guesses for reaching a B2-C1 level being at A2-B1 level(based on my experience with Japanese):

  • distant Indo-European languages: 300 * 2.5 =750 h
  • distant Non-Indo-European languages: 300 * 4 or 5 =1200 - 1500 h

In sum:

  • The Minimum Effective SLA dose for closer L2s should be feasible for the vast majority of L2 learners. If 2 Pomodoro blocks are too much per day, just choose 1 block and double the number of days (2 *200 days = 400 days)
  • The middle category (i.e., distant Indo-European languages) should still be feasible for many L2 learners: ca. 2 years with 2 Pomodoro blocks per day.
  • The distant Non-Indo-European languages - well, they’re simply “beasts” where the Minimum Effective SLA Dose per day is probably 2 - 3 h. And that matches roughly Florian’s (bamboozled) daily learning time in Chinese.

@xxdb
If you find something more efficient (or interesting), I’d love to hear about it - as always :slight_smile:

2 Likes

“you are getting the collocation exposure for those precise words that you don’t know.”
In my opinion, collocation exposure is good, but not enough.

That’s what I like, for ex., about ESLPod.com or Mercedes’ Spanish Podcast from Barcelona. They provide background information about the collocations (including idioms):

  • What’s the language register: formal -neutral - informal - slang?
  • Are they outdated or not?
  • Are they used in oral or written forms?
  • Are they only used in certain contexts (sociolects)?
  • Are they only used in jest?
    etc.
    I’ve even specialized literature with tens of thousands of such collocations - unfortunately, as books. And that’s no longer state of the SLA art.

However, even SEs specialized in collocations are not the right tools for language learners…

1 Like

Awesome. Thanks for the stats.
I guess what you’re saying is I’m too impatient.

1 Like

Slow clap my friend.

1 Like

I understand your possible “impatience” all too well (at least, when it’s a question of temperament :slight_smile: ).

However,“Russian” is not Dutch, for example.

Dutch feels like a German “dialect” to me because there are so many words I can easily guess. Therefore, it’s a pure pleasure to surf the Dutch text waves.
In Japanese, in contrast, it feels like I have to wade through an SLA swamp - with no end in sight. There is progress, of course, but often it doesn’t feel like it…

In short, there are simply L2s that are intrinsically more difficult than others. And Russian is definitely one of them - at least for us coming from the Germanic language family.

But that’s hardly any news to you so I’ll stop preaching to the choir :slight_smile:

Re ChatGPT:
I think I need to figure out good use cases for it:
I’m already drowning in dialogues, so I don’t need more AI-generated dialogues.
However, chatting with generative AIs could be a nice addition to my study routines…

1 Like

@xxdb
Re the “Minimum Effective Dose” for learning and maintaining an L2 (see your other recent thread)

I think your idea reg MED applied to SLA is excellent!
Unfortunately, I only found the usual “stuff” in yesterday’s Google search (i.e., Tim Ferriss and his “apple sentences” to deconstruct any L2, the forgetting curve, spaced repetition and active recall, etc.).

What would be more interesting is something like this applied to the brain / mind:

So we’re left with our own language learning experiences and intuition…

1 Like

I’m glad you bring up the fact that there is a difference between input skills (reading/listening) and output skills (writing/speaking). They are very different skills and likely use different parts of the brain.

I’m going to bring up a different kind of learning that I think applies to language learning. Morse Code. I happen (after 66 years of ham radio) to be a very skilled Morse Code operator. I can receive (an input skill) at easily 40 wpm and with effort up to 55 wpm. During the telegraphy heyday, many professionals could handle those speeds and more, but the workaday telegraphers tended to stick around 20 wpm. There are reasons for this.

The way we learn Morse Code is to first memorize the pattern of dots and dashes for each letter of the alphabet. An A is dot dash (sounds like di-dah), a K is dah-di-dah and so forth. E, I, S and H are dit, didit, dididit, didididit respectively. Our pattern seeking brains pick up these repetitive patterns and use them to memorize them, but it’s a trap.

When we listen to code, we look for these patterns, almost like a lookup table, but not really. We hear didah and think A. And we do that for each letter. But that takes a LOT of processing power which takes time. It’s done by the conscious mind which is slow, lazy, easily bored, easily distracted and forgetful. So the speed at which you can receive Morse Code this way is limited to around 13 wpm. With lots of practice, some can get that to 20 wpm.

My point so far is that language learning is the same. We memorize words. We memorize word patterns (grammar), We memorize common phrases. We are depending on the slow, lazy, forgetful conscious brain to do all this. When a word comes at us, we have to consciously recall it. We have to remember what the previous word(s) were, and we have to construct a sentence of these words in our head. If we can’t recall a word, we stumble and miss the next word and quickly get lost in word soup.

At this point memorization and recall are far too slow and we are very limited just like in Morse Code, because we are relying on an inherently slow processor.

In Morse Code, the break through comes when we stop listening. In order to learn to receive/input faster, we have to stop trying to decode the patterns. With hours and hours of input, at some point, the task gets handed over to the subconscious mind with is very fast and operates in the background. We can’t conscioously pass it onto the subconsious. We have no real control over the subconscious. Soon the subconscious brain starts seeing patterns of entire words and simple phrases instead of individual letters or dits and dahs. We understand what is being sent without consciously trying to, and at higher and higher speeds. Once that ~20 wpm barrier is breached, it’s only a matter of more and more practice to double that speed to 40 wpm.

Again, I think my language study in Norwegian follows a similar path. I don’t try to memorize anything. I don’t study grammar except to “break the code” about why some word order seems strange to me.

My reading has gone from word by word to largely sight reading where I often no longer see the individual words. The conscious mind isn’t fast enough to reach that level of reading. The faster subconsious is coming into play just as it does with my Morse Code.

I’m completely “fluent” in Morse Code. I can converse with someone across the world at 40 wpm and at the same time answer questions from a visitor to my station. It used to drive my mom nuts when she would ask me to take out the trash or clean up my room while I continued my conversation via Morse with someone on another continent.

I’m not suggesting any shortcuts, nor am I suggesting any plan of study. I’m just offering a reason why just about everything seems to work. By engaging the conscious mind and overloading it, sooner or later it will pass on the tasks to the subconsious. I don’t think there is anything the conscious mind can do (flash cards, LingQ, Duolingo, etc.) to speed this up. That breakthrough will happen when it happens. We can manipulate certain aspects of learning with “shortcuts”, But overall, I question whether they are breakthroughs or just parallel paths.

Frankly, I don’t believe there have been many breakthroughs in language learning since the Rosetta Stone, and I’m suspicious of any claims to the contrary. It’s all being done with the same wetware we’ve had for a couple of hundred thousand years, and have yet to fully understand.

8 Likes

@xxdb

“Before I read the book”

You are reading a book?

At an intermediate stage, my focus is very clearly on spoken language. Spoken language uses much less vocabulary, so I choose to learn this vocabulary first before that of books. I see it essentially as an intermediary goal. The Paul Nation paper said in English you need about 6-7k word families for a 98% coverage of the vocabulary in a wide range of spoken language topics compared to 8-9k for a wide range of written texts. The exact numbers obviously aren’t transferable to Italian or Russian, but it gives you an idea that maybe there’s a 50% increase in vocabulary needed to understand written language after you’ve learnt the vocabulary for spoken language. In other words, why learn the 9th 1,000 word family before the 7th 1,000 word family? That’s the way I see it. For this reason, I am very selective in my material (no books yet). Though, maybe you have already acquired these more frequent word families already, so that’s why you’re reading books. If so, you are at the upper intermediate already. Or maybe in Russian this spoken-written vocabulary divide doesn’t exist? I don’t know.

EDIT: I think we talked about this before and you called it ‘laddering up’? I consider understanding podcasts and YouTube videos to be a lower rung than understanding books. ‘Laddering up’ in difficulty of material really does increase ‘efficiency’ and you will feel like you are making more progress because you actually are.

3 Likes

Thank you sir. That was a beautiful story, thanks for sharing: I really enjoyed it.

Substantially I think you’re right. I’ve experienced the conscious->subconscious shift when I learned Spanish. I think what happens is that you build a secondary language center which is eventually accessible by the subconscious mind. What it feels like to me when I hear Spanish is it just “hits me” and I just know what it means. And I don’t mean I know what it means in English. I know what the direct meaning of it is in Spanish without having to translate it.

With French and with Russian I’m not there yet. I have “glimpses” of it with Russian where there are e.g. 3-4 words “run” where I “just know it”.

That said, I do want to quibble with your theory that there is nothing new under the sun with respect to language learning. SRS (the basis for anki) is new and it works. Now it doesn’t work for everybody because it is BORING. But it definitely works.
That said, my experience so far has been that it works to take you to high-beginner/low intermediate ONLY. Then you need something else. With Spanish I went whole hog old-school and immersed myself in the latin community in Toronto and after a year or so (after the initial six months just drilling SRS) I would say I hit the early stage of fluency.
With French I never became fluent because I didn’t really care. I was doing an experiment to see if I could learn another language other than Spanish. It worked. I got to intermediate. My mistake, though was I forgot that I didn’t use anki for Spanish once I had gotten past the first six months. So I’m inventing or trying to invent something that will get me some kind of similar results to my Spanish experience but without having to hang out with Russians for 5-10 hours a week. Not that I have any issues with Russians in particular (except Vlad the evil, screw that guy) but I just don’t have time to immerse. So I’ve been looking for a shortcut to go to advanced. anki is not it and anki plus lingQ is not it. At least in the basic form.

However, I think anki+chatgpt+lingQ might be it. We’ll see.
Anyhow, good chat. And thanks again for sharing your story.

3 Likes

@nfera:
You are reading a book?
At an intermediate stage, my focus is very clearly on spoken language. Spoken language uses much less vocabulary, so I choose to learn this vocabulary first before that of books. I see it essentially as an intermediary goal.
Yes. My method is the opposite of yours it seems. I start with spoken material and don’t even try to read at all. I kept that up until the very last minute when I did French two and a half years ago. This time around I decided to give lingQ a full hog try and have been doing a combo of lingQ reading plus anki words (audio) for the last 8 months or so.
My experience has been that I’m improving. It feels very slow but there is definite progress. Also, wierdly, my spoken comprehension is improving also even though I am barely watching any videos. I feel like, however, that there is merit to what you say: that my spoken comprehension might have levelled up even faster if I had stuck with watching videos. Reason being that video was right at the edge of my comprehensible input window whereas reading was essentially a bit above. So maybe reading plus anki in hindsight might have been less efficient than continuing with my standard practice of anki audio plus watching videos. That said, I can read now, whereas 8 months ago I definitely could not. So there’s that. I guess I can count that as a massive win which I have not really been considering so yeah.

Anyhow yeah I love these chats with other enthusiasts to see what we can come up with as we riff on each other’s methods. Thanks for your input.

2 Likes