Learning Persian with Anki, ChatGPT and YouTube
cjauvin.github.io
cjauvin.github.io
I studied Dari own my own and at college as an elective, and ended up taking a job with the ICRC to investigate ISAF war crimes in Afghanistan right after I graduated
These days Dari is my most comfortable second language (and I have quite a few of them)
I'm not sure if, given I had to start from scratch again, I'd go down this route - the description and screenshots seem very overstimulating for me
The most important parts of my language learning in Dari (and Pashto) - the "aha" moments if you will, were trying to express something, making a fool of myself, making everyone around me laugh, and then being gently corrected in a long-winded way (usually because I couldn't understand a simpler, more direct correction)
In hindsight this feels like a very equitable cultural exchange - I learned something valuable about the language and culture while giving my interlocutors a funny memory to share with their friends and family
As an Iranian-American, I've made similar blunders in trying to communicate with my Mexican friends and colleagues. "Hey amigo, do you need a chauqeta?" I would ask to the eruption of laughter. If you look up "Chaqueta", translation services will happily tell you it means "Jacket". Don't trust it. Apparently to some Mexicans it also means jerking off.
I know that some cultures consider it rude to laugh at this sort of things when it's not the speaker's native language (brits don't even correct me when I misuse/misspeak english because they think it's rude), but for many (most?) cultures it's really not, that's not being a jerk
This is a regional bro distinction. I have certainly heard chaqueta used by respectable people in Jalisco. This joke is a reflection of people that don't travel.
In Jalisco y Nayarit the normative word for the large beer bottles is caguama. In central MX that word is ghetto.
What happened for me was that the ICRC was so desperate for Pashto speakers in Afghanistan for this sort of work (they weren't allowed to hire people for the war crimes investigation work from Pakistan or Afghanistan due to neutrality stuff) that for a time they ended up accepting people who had learned Dari (ie. demonstrated the ability to learn) and gave them 3 months in-country to work with a Pashto teacher and learn enough to be able to be functional in their jobs
This was right after the financial crisis, I had just graduated with a liberal arts degree and had no other job prospects, and I had an interest in Afghanistan, so being paid to learn Pashto and possibly having a job afterwards seemed like a very lucky break to me at the time
One app I used early on when beginning French was Clozemaster, set to keyboard input (instead of multiple choice). The largest benefit was I didn't have to make all the decks, they progress you through the most common words (used in context), and there are ChatGPT grammar explanations for everything if you wanted to drill into it. It sounds very similar to what OP created for themself.
At a certain point you just need to switch to native content, but at the beginning I found Assimil + Clozemaster + comprehensible input on YouTube to be able to get me to watching regular French TV in maybe 6 months.
It sucks though, it's also the one thing that makes me constantly not be consistent using Anki, I get tired of creating cards and stop for a while.
People are just repeating this advice about making your own decks, and it's based in nothing but having had it repeated to them. Spaced repetition is boiling in pseudoscience and ancient studies that don't say much other than that there's a forgetting curve.
Most people are just parroting stuff they read on the Supermemo wiki (or somebody read off the Supermemo wiki and repeated to them like they came up with it), and all of that is just thoughts off the top of one guy's head. His innovation is that he wrote a program to do Leitner boxes before he had ever heard of a Leitner box, but people treat every word like gospel.
The only five things I can say for language learning is to go really hard on systems in a new language that are completely unknown to you (like Romance conjugations for an English speaker); only drill sentences, not individual words; always say your Anki answers out loud, and read out loud as much as you can; comic books have pictures, too; and once you get comfortable in an L2->L2 dictionary, you're a more comfortable reader than a lot of natives.
-----
* Random Anki decks for a few European languages: https://sookocheff.com/post/language/cloze-deletions/
(Edit: the lovely thing about 10K algorithmically generated clozes is that they're utterly disposable, unlike cards that you make yourself. If one is a leech, forget about it. You'll see another one just like it when you get to the point that it won't be a leech for you.)
* Instructions on how to generate your own in other languages, for developers: https://sookocheff.com/post/language/bulk-generating-cloze-d...
(You could probably point out the above URL to an LLM and it would generate the code for you.)
* Anki to learn Romance conjugations first: https://www.asiteaboutnothing.net/w_ultimate_spanish_conjuga...
I feel even better than natives sometimes because they learn conjugations in order at school, and when asked to recall them out of order (or hop from form to form) get confused. Once you have conjugations, you can read anything with a dictionary (and the online dle is the best dictionary I've ever used.)
I'm about to start again with KOFI French, but I had to do a lot of work to get my mouth and ears adjusted to hearing French as anything other then murmurs, and to be able to read (luckily for me, French is the opposite of perl and read-only instead of write-only.) There's a lot of stuff in French I want to read; reading all of the stuff translated into Spanish from French (but never into English) has got my beak wet.
Also, Spanish-language comic books will make you forget about English-language comic books. And they are very online, examples: http://columberos.blogspot.com/ and https://comicsmexicanosdejediskater.blogspot.com/
Also lots of other good material for vos: https://ahira.com.ar/ including https://ahira.com.ar/revistas/skorpio/ and https://ahira.com.ar/revistas/hora-cero-suplemento-semanal/ which is a landmark of literature that is too good for us (El Eternauta spans the entire length of the series.)
For hearing French, I went through the old FSI French phonology course at the beginning, as well as all the grouped (A1/A2/etc) comprehensible input from this channel: https://www.youtube.com/@FrenchComprehensibleInput/playlists. Oh, and I did a bunch of the French listening comprehension on Yabla for a few months: https://french.yabla.com/.
I did the same. Got me interested in the old programmatic methods and how they could be adapted to LLMs.
Also, pssst... https://seulementbd.blogspot.com/
> using preselected word lists makes sense
I have never used Anki for word to word translation, because there is no Spanish word that means any English word. You've restricted yourself to a particular paradigm from the beginning. I mean, do whatever you need to do to get a foothold, but you want to get away from L1 as much as possible as soon as you can.
I know you consider it pseudoscience to force creating your own cards, but I do find premade decks result in a lot more leeches. How do you avoid them when relying on these thousands of auto-generated cards?
Also, one of the many parts of the spaced repetition lore that I do agree with is that if you keep getting the same card wrong in the same way over and over again, you're building up a weird habit that is going to be tough to break. Better to trash it. [edit: you can't do this with the conjugation cards, though. If you keep failing a particular conjugation card, you need to stop, write it down, and spend time with it individually. All of those are important, except maybe the unique conjugations of europeizar.]
> if you're still in the earlier stages it's much more fun to learn a variety of words and focus on the most common ones.
I believe of course in the have fun rule above all others, because this is an ultramarathon, not a sprint. You're going to have to get enjoyment from the process if you're going to stick to it at all. I get a "dopamine hit" every time I get a card right.
If you wanted to sprint, the best way is probably doing the full old-school* Glossika method where you go over the day's sentences, you listen and repeat, you listen and transcribe, then you repeat on your own and record. The next day you start by listening to your recordings and figuring out how to improve them, rerecord, go over your new sentences, rinse and repeat for 3-4 hours a day. You can certainly hammer a language into your head that way, but you probably need a tiger mom threatening to withhold food or something to keep you doing that for 6 months.
About creating your own cards - I've done thousands (not language related), and I've learned and remembered things with them, but there's no science behind writing a good card. Everybody is on their own and flailing, and asking themselves "what would Woźniak do?" rather than coming up with formal rules and testing them. I've got ideas, and there are a few datasets (of people doing spaced repetition sessions over time) available, but I think that treating cards as a black box, discovering the relationships between them over time (through finding which cards are passed and failed together), and creating some sort of internal market of cards between adversarial LLMs could generate good decks, and generate mappings that would allow failing or passing one card to affect many other cards rather than the single card alone.
The best thing about FSRS to me is the decision to stop asking the user about process, and just ask them about goals. Some of the Anki community seem on the verge of giving up on the self-grading, too, and just moving to pass/fail, which I couldn't support more. I never used anything but pass/fail, with the settings from refold (https://refold.la/simplified/) which have been obsoleted by FSRS, which works better with pass/fail only. As I said before, a lot of how these things were designed from the beginning was simply cargo-culting Woźniak.
----
[*] I haven't tried the new school online version, but I'm sure they have varieties in the same mode.
I think what's key is that I'm taking the words and conjugation rules I'm learning and using them relatively quickly, often that day or that week. I.E., I'll come across words in Anki, then hear them in a baseball broadcast or see them in a news article. Or I'll recognize what tense something is because of the rules.
So it's supplemental, and maybe that's why it's sticking better. I don't think I'd want to create decks constantly, I created one 140 card deck and that was enough.
Finally, I do frequently use memory tricks to create associations so maybe my experience with memory castles, mnemonics, and other techniques (which I use on cards I forget frequently to create links I'm unable to create quickly on my own (or to differentiate similar words (or words that are the same but in different tenses))).
Yea, it's _work_!
It's gotten quite popular enough that I've gone full-time on it
https://ankiweb.net/shared/info/1531888719
FWIW I did get a lot more mileage from building my own deck vs a custom deck too, would recommend that approach regardless once you're past the initial vocab bootstrapping phrase.
I don't know what similar tools exist for learning other languages but it does help a lot for Chinese.
Now I just click any word or phrase when browsing the web, and it shows me the exact meaning in context, thanks to an AI-powered dictionary that works for any language. I save the word along with the sentence, add a quick note if I want, and later export everything straight to Anki.
It’s cut out all the friction of card creation for me. I use it daily for English and Chinese, and it’s made sticking with Anki sustainable again.
Anki is probably my most beneficial single tool. Though if I were to do it over again I'd follow more or less the poster's strategy. Maybe 80% comprehensible input for listening and 20% Anki for vocab building. At least until I could watch native TV without much effort. I've played around a bit with LLMs, but still haven't found a really great use case for my study.
On the otherhand I think consistent practice (with growing difficulty) trumps technique. Whatever process keeps you motivated to practice month after month is most important.
I love anki and use it for Spanish which is showing marked improvement. I do vocab and conjugation with Anki.
Then I just find other ways to immerse myself and call it a day.
- Spanish audio for sports whenever possible - Interfaces for personal computers/devices - Picking up the Spanish language weekly from the little box on the corner - Listening to Spanish artists - Reading the news in Spanish instead of English (One major benefit here is consuming far less news) - Writing notes for work and personal projects - Texting friends
It all really adds up over time and is definitely doable even as an adult, but it requires a ton of work, so being able to find ways to incorporate it into the activities I'm already doing is key for me on top of the more active Anki learning.
GenAI also been a big helper when I run out of content. "Write me an essay involving [subject I want to learn about]. In my response after reading, any word I've written separated by a comma generate a CSV of the format "that word, english definiton"." I'll then just dump those new words into Anki.
I dumped my 400 hardest recurring anki words in it and listen to the stream whenever doing chores or driving. Then sync with my deck again after a while.
Can you help me out and give it a try, you seem like the target audience and i'd value your feedback. If your target language is not available or want to upload an anki deck I can help you out.
If you find it useful, you can register for free and create new streams on any subject. Send me a mail on alex@longyan.io if you'd like more stream/content quota or if you want to try the Anki thing, I'll gladly set it up for you.
This is a fine way to bring the material into your "cache" but you aren't doing the work required to learn a language: Communication!
The approach presented in the post is not necessarily on conflict with that, but it seems like extremely laborious preparation stage. And despite expectations, learning the different alphabet is usually unnecessary until fairly late into the learning process.
I was super bullish about ChatGPT's Voice Mode, but it is so eager to respond that I never get a chance to complete sentences!
Edit: it looks like they updated the Voice Mode UI since I last used it - hopefully they retained this capability.
You also didn't understand that cards in anki can have more than 2 sides. Making Persian writing->Latin transcription then Latin transcription->English translation is a huge antipattern, when you can have all 3 on one note (simply add a 3rd field, also there's a built in "hint" field) - and above all should not use a Latin transcription at all (Notably, in the deck settings, you can generate cards from notes in different ways.)
هیچ کُدام now has the o marked, that easy! (N.b. author, another issue with your method is... Youtube videos are teaching you random things without structure. Colloquial Tehrani Persian turns án/ám into un/um which you are learning in your vocabulary. But you can simply learn the replacement rules and apply them when speaking in certain contexts.) Please use a good textbook instead. In 100-200 hours, you should be around B2 with a good program. (Better Assimil courses bring that down to ~75 hours.)
I strongly recommend:
- Baizoyev & Howard’s Beginners Guide to Tajiki - teaches the written language, with all vowels marked, and multiple dialects, this is by far the fastest way to master Persian. Reading/writing in Persian script is essentially mechanical with a good base in the language and not an issue, but you can read all Persian classics in the Tajik script with all vowels marked...
- Lambton's Persian Grammar - teaches the written languages along with colloquial Iranian usage
- Elwell-Sutton's Colloquial Persian - uses Latin transcription, quickly teaches the grammar and a nice vocabulary
-----
But going further, if a vowel's not marked but feels necessary:
> In 1792, Edward Moises already suggested not trying and just saying e
Different dialects differ a lot on short vowel usage (even in grammatical forms), so this is a surprisingly valid trick.
How do you know they are not learning to type?
> you complain about letters having different forms
Where did they "complain"?
The OP's article:
> From this, I will extract three screenshots (with the MacOS screenshot tool). First, to create a card of type “basic” (one side). I use this type of card to exercise my reading, which is very difficult and remains stubbornly slow, even though I know the 32 letters of the Persian alphabet quite well by now. But the different ways of writing them (which varies by their position in the word) and the fact that the vowels are not present makes it an enduringly challenging task.
It doesn't sound like they literally can't type in Persian, or they're complaining about how it's written, at all. They're merely stating the fact it's difficult for them (like every language learner).
They also screenshot the English part too. So presumably they screenshot because it's faster, not that they can't type.
> Author, you're not properly engaging with the language
Strangely condescending. They're focusing on reading and listening, which is legit for beginners.
I do agree that the use of Anki cards is suboptimal though.
He's talking about reading as a challenge. Not typing. It's very clear and unambiguous from the original article.
I'm flagging your comment for claiming I didn't read the article. If the author has trouble reading letters, how can he type well (which was your first point)? Addressing that first would prevent him from using transcriptions and dual sets of flashcards.
At minimum, consider a few hours now compared to the time saved by halving the number of flashcards you need.
So it seems to make a lot of sense to learn with that aid and later transition to no vowel markings (or reduced / the normal amount)
EDIT: to be more specific, language like "You complain" and "You also didn't understand" can appear abrasive and scornful. Removing them would probably make the effect you're hoping for (proper learning techniques for the language) more meaningful.
It’s important to be a bit strong willed against this type of comment in the internet and stick to your instincts. Not to say good things cannot be learned out from others, but finding out what works for your particular needs through self experiment is really important, and on the long run time exposure to the target language trumps over methods.
Some examples:
- Learn grammar from a good book vs. absorb it naturally from material.
- The [tool I’m using] is better than [tool I’ve never tried].
- The [method I used for a month] is better than [old method I used for years].
- The [tool I’m selling] is better than [my competitor’s].
It’s easy to get discouraged if you don’t see the patterns and think you are doing something wrong, it’s better to keep vigilant.
Helping with language learning is one of the things I think ChatGPT is excellent for. I have a long-term conversation only about Korean, and I can ask questions like "how would a Korean understand [some grammatical structure]?" and it gives very insightful answers, and even refers back to vocabulary that I've already used or other discussions about similar topics.
And in the end, in English it should be "Persian" and not "Farsi", that is where the actual move should be. How sad and historically wasteful if we started to do that to all languages, "deutsch", "zhongwen" or "elliniki" instead of German, Chinese and Greek
This is not true.
It happened after the 1979 Iranian revolution, when Iranians abroad wanted to call it Farsi out of cultural pride, using the same word in their own language, rather than Persian which is the "foreign" word for it (from Greek/Latin). It was literally reclaiming the name. Then the media followed suit out of respect. It was cultural sensitivity.
Today some non-Iranians and therefore groups like the UN prefer "Persian" because variants are also spoken in Afghanistan and Tajikstan, and Farsi is a reference to the Fars province of Iran, so Persian can be seen as more neutral. But then again, not many people complain about "English" being associated with England and not being neutral enough to Americans or Indians. So it's definitely complicated. But it's also definitely not about trying to diminish anybody's "prestige".
I think it just was a random occurrence / perfect storm in the US where some Persian speakers moving there didn't know there was an English word for it, and where the local population were more used to hearing names of unknown languages, and it happened to stick around and start spreading. And _then_ it maybe became a cultural pride or whatever thing with media following suit like you said
The difference between Persian and Farsi matters in english world because it is political. In english, Farsi doesn’t carry prestige, same way iran doesn’t carry Persia. But they do for those in iran. I’m not familiar enough with afghanistan to know first hand, but perhaps same can be said for Dari. I might be wrong though on this last one.
My organic chemistry is... terrible to say the least. I might try Anki again if you say so!
My method is more primitive, I first get a simple overview of the topic (LLMs are great at this). Once I have a feel , i flick through the material book/paper highlight important info that stands out or info that I want to remember and personally for me, Im not trying to understand things as I highlight, once I'm done a chapter or a big section, I pull out my anki and start making questions against the highlighted parts.
When Im making questions, usually I make one questions that corresponds directly and I use the highlighted part as the answer with minimum change expect for readability and then I make several other questions that takes different parts of the highlighted answer, so that I can have an almost lego like breakdown of questions that can help me recall the "bigger question", also I make sure the questions arent to direct and force my brain to think and retrieve the answer
I hope this helps, this is the article that inspired me to read this way: https://augmentingcognition.com/ltm.html
I see some individuals creating LLM generations of the cards which is definitely helpful at times, but also a double edge sword when it comes to learning. But I don't see anything taking in the users voice as an input which would be immensely beneficial
Manchmal, wenn ich ein Wort noch nie gesehen habe, kann ich die Bedeutung herausfinden, weil ich das entsprechende Wort auf Dänisch kenne.
People do underestimate how much a good teacher and putting in the work can make a difference. Money well invested if you actually NEED to use the language and will have opportunity to.
The flexibility of being able to dynamically generate a ton of Anki cards using a script is trumped by just using ChatGPT to generate and grade answers. This will not work well for advanced language learners but for German it works really well - as much repetition as you need to master specific skills - up through intermediate level.
1. The lack of a good spaced repetition algorithm.
2. You'll end up with an order of magnitude fewer flash cards. When it's easy to copy/paste, there's no way you can create all the flashcards you need by hand in a decent amount of time.
> The flexibility of being able to dynamically generate a ton of Anki cards using a script is trumped by just using ChatGPT ...
I would never advocate doing it with a script. All my flashcards were created by me - either by typing or selective copy/paste.
Part of it, for me, is that “magnitude fewer” is a benefit, not a drawback. Mastering a bounded scope and then intentionally expanding it by hand writing new cards keeps me focused on the rock face that I need to practice and internalize. Having the next 1000 words be two clicks away, and then dozens more decks waiting after that, makes the whole enterprise seem incredibly daunting.
That may not be your experience, of course, and I wish Anki were effective for me.
Most of my cards are manually typed - not copy/pasted. And yes, it totally sucks, so I don't produce as many.
I've been doing it since 2018, and the total number of cards I have is a few thousand - probably under 5000. If you review daily, you'll often end up with days where you have nothing to review. Very manageable.
I recently got back to coding in C++. I'd forgotten some aspects of the language, and also needed to come up to speed with many of the newer features in C++23 (I was mostly an 03 and a bit of 11 person).
It's a lot of cards to create! I'm doing it slowly, but this would be totally infeasible if I had to do it by hand. Still mostly manual typing, although I do copy/paste code snippets.