How not to learn Spanish: Building too much stuff, not studying enough

[emk]

I have a real language log. If you want to find fun stuff to do in French, you might find some useful information there!

This is not that log.

This log's goal is to maybe learn a tiny bit more Spanish. Sometime in the next, oh, 30 years. I intend to go about this in the absolutely worst way possible, which is inspired by these xkcd cartoons:





Instead of, oh, studying, I will build tools to help me study. Instead of sticking with the project consistently, I will likely completely forget the project for years at a time. Do not be like me.

The story up until now. My previous work on Spanish was inspired by Sprachprofi, and it involved audio flash cards and writing substudy. This worked out surprisingly well:

Next Tuesday will be the 60-day mark for this experiment! My deck now contains 437 mature and 263 recent cards (and 271 suspended). I've done 20 hours of Anki reviews, and 3,483 reps. I've learned 3 episodes of Avatar.

While waiting for some software to recompile, I watched parts of two more episodes, all without subtitles. Here are the results:

Episode 5 (above, studied using subs2srs). Comprehension is very high—about where I was after several seasons of Buffy, and after passing my B2 exam. This is pretty remarkable, when you think about it—subs2srs has allowed me to actually "learn" three episodes to the point where I have essentially B2 comprehension. But only for those three episodes.

Episode 6 (prepared for subs2srs, but not yet studied). This case is a bit more interesting. I've recently watched this episode and spent some time aligning the subtitles against the Spanish audio. So I have a pretty good idea what's going on. Overall, this gives a nice boost, especially in the scenes I've watched several times. I can follow the plot, and I can sometimes get big sections of the dialog.

Episode 7 (watched in French about a year ago). This is the purest test of comprehension. I remember the plot of this episode (more or less), but I've forgotten all the details. And I have good news: I can follow most of what's going on! I'm definitely missing over half the dialog, but I almost always get at least several words per sentence, and sometimes I get multiple consecutive lines. Overall, this is actually pretty fun.

[...image lost to history...]

So it looks like I'm definitely nearing the point where I could start just watching Avatar for fun. Even with my strong background in French, this is pretty awesome—20 hours of "official" reviews, according to AnkiDroid, and I'm more-or-less able to follow an actual TV series (with lots of guesswork). (For comparison's sake, Sprachprofi was watching a Japanese TV series unaided after ~30 hours of Anki, with 50% comprehension, but she's a talented polyglot.)

Personally, I'm just happy that my brain is starting to understand things like hay que hacerlo automatically. My English-and-French discount doesn't actually help much with phrases like that, but after 3,483 reps, they tend to jump right out at me.

[...time passes...]

Episodes 11 & 12. Ah, that's better! My comprehension was much better than it was with episodes 9 and 10. To review:

Episode 5 (studied with subs2srs): 80+% comprehension.
Episode 6 (reviewed without subs2srs): Variable, but good overall.
Episodes 7 & 8: Less than 50% comprehension, but I could follow the plot pretty well! Definitely fun.
Episodes 9 & 10: Definitely harder than 7 and 8. Rough going overall.
Episode 11: Not as good as 7&8, but definitely better than 9 and 10.
Episode 12: Wow, this was great! I followed almost all the story, and I understood some sections solidly.

Restarting the project. I have just dug up the hard drive with all my fancy bilingual subtitles, and I am currently rewatching Avatar season 1, episode 1.

avatar-no-se-escaparán.jpg

The remarkable part is that a half-decade of total neglect of my sub-A1 Spanish, I can actually still understand most of the dialog so far. This is a bit startling—I would expect a sub-A1 language to decay almost totally in 5+ years. But I came very close to memorizing this episode. The lines of dialog are stuck in my head like a song lyric from high school.

I also spent yesterday evening mocking up an SRS UI that does nothing, just to try out some new tools.

srs-ui-mockup-small.png

Button labels? I am thinking a lot about button labels right now. Anki is designed for learning cards. But when you can create 500 cards with almost zero effort, you want your SRS workflow to change. Any individual card is completely expendable—if you actually need to learn something, you'll see it on another card right away. Instead of failing cards, just review them slightly more often, or throw them out. If a card makes you say, "Ugh," delete it immediately.

In this paradigm, you're not really learning cards. You're "concentrating" them, I guess? You just want to make interesting input reappear more often in the hope that it gets stuck in your head automatically. Less stress, more fun.

Also, I review other people's tools. Some of these are really nice. You should try them, instead of waiting 30 years for me to maybe ship something! (I can actually ship software quickly, if it's putting food on the table! But again, that is not the goal of this project. )

[emk]

There was a moment there, for maybe 20 seconds, when I was just watching the episode and understanding it, before realizing, "Oh, yeah, this is in Spanish." So happy.

I'm relying pretty heavily on the Spanish subs, and I'm definitely working hard to dredge up a bunch of stuff from memory. But I'm doing it in real time. And I think I'm slowly relearning most of my Spanish verb conjugations from context.

[iguanamon]

Yep, "do what I say, not what I do" . Of course, you know that if you did "Assimil L'espagnol" alongside substudy, you'd probably be learning on steroids- much, much faster. Still, I get it. I've always said you were onto a winner with this project.

As we say here on the forum, what you learned was still with you. You just had to wake it up. Looks to be a very successful method. I wish some programmer with a lot more time on their hands than you have could make it more easily available to the masses.

Now, five years hence, the phone is becoming more and more the primary computing resource for most folks. Any program for the masses would have to be phone friendly. The copyright problem would have to be solved... dvd's are a dying medium. Details, just details.

Just wanted to say, I'm sorry the forum had its access problems a few months ago... but it brought your voice back to the forum. It has been sorely missed. Looking forward to seeing more of your progress.

[tastyonions]

Best of luck on the not studying yet still hopefully making some progress.

[emk]

Yep, "do what I say, not what I do" . Of course, you know that if you did "Assimil L'espagnol" alongside substudy, you'd probably be learning on steroids- much, much faster. Still, I get it. I've always said you were onto a winner with this project.

It's hard to judge precisely, because English+French gives a ridiculous discount on Spanish. But my gut instinct says Subs2SRS-style cards outperform the Assimil active wave pretty noticeably. Assimil l'espagnol is painfully slow compared to Avatar. Like, I tried to do it, and it just killed me. So.... slow....

Realistically, I was most of the way to B2 in French before I could just casually watch Avatar extensively. Sure, it's a lot easier than something like Y tu mamá también, which feels equivalent to the French movies I could finally start tackling closer to C1. So Assimil is stuck using A1 audio, but I could literally start using B2 audio before I could conjugate a Spanish verb.

And this is interesting, because it reminds me of the "Farnsworth method" of learning Morse code:

The classical way to learn [Morse] code is to start slow and then build up to a higher speed. The problem is that you tend to develop a "table lookup" in your brain, and a plateau develops as you try to progress beyond 10 or so words per minute. You just can't "look up" the characters quickly enough. You need to be learning the characters by sound and not as a pattern of dots and dashes. Do this from the start and you won't have to relearn it as your speed improves. Writing them down should become a reflex.

Most experts agree that the Farnsworth method is the best way to learn code at a given speed. With the Farnsworth method individual characters are sent at the target speed, but extra space is sent between characters and words to slow the rate at which you have to translate. As you improve the extra space is decreased. This way you learn from the beginning how each character sounds at the target rate. Translation becomes more of a reflex.

Learning slowly requires you to put in a second big effort later to speed up. But what if we could learn at full speed in the first place? And this is where I think Assimil has missed a trick. If you gave me access to their team and to their e-méthode development budget, I'd build an integrated SRS and re-record some of their courses with much faster and more natural audio.

Now, the interesting question is the active wave. I have some wild hypoteheses about how to replace this, using a mix of cloze cards and L1 text → L2 audio cards. The trick would be to reuse the original Subs2SRS audio cards, the ones stuck in my head like earworms. Could I learn to speak by pulling in earworm fragments and re-using them as speaking "chunks"? I don't know. But I'd love to try.

The other weakness of this method is that it leaves you with good raw listening skills, and even some grammar instincts. But it leaves you with almost no vocabulary. So once you've got the infrastructure to do audio SRS, it's time to invest in better SRS for web pages, too. Which Language Reactor is apparently doing.

And I love the way Migaku is using ChatGPT to "explain this word in context" on the backs of cards. Sometimes ChatGPT gets it wrong, but it's still more useful than digging around in an online dictionary. To make a lot of these SRS-heavy methods work, you need to carefully eliminate friction at every step of the process. Watch/read native media. Click a button to capture. Later, spend a couple of minutes turning that captured input into cards, with as little time and effort as humanly possible. Finally, have the cards automatically pop up in your SRS, with no import/export nonsense. Migaku is working very hard at all of this, actually, even if their stuff is clunky.

Now, five years hence, the phone is becoming more and more the primary computing resource for most folks. Any program for the masses would have to be phone friendly. The copyright problem would have to be solved... dvd's are a dying medium. Details, just details.

I think the only way to actually make it work for a mass audience is what Migaku is trying: Use a browser plugin to capture short audio clips and image stills from online videos. But this assumes a hybrid workflow: You watch videos on a laptop, and review cards on your phone. If you go 100% phone, you'd need to develop your own courses from the ground up. Phone browsers don't support plugins.

Looking forward to seeing more of your progress.

It's great to see so many familiar faces again!

[emk]

Recovering Spanish study materials from my backup server. I have a 5+ year old Synology backup server. This was a nice little machine, but it hasn't aged well. Among many other things, I used it to store language-learning materials (like subtitle data and processed OPUS data) that didn't fit on my then-tiny laptop SSD drive. I did this using the incredibly geeky tool git-annex, which is designed to allow fine-grained archiving of many large files. If I want a file, I can ask git-annex to find it, and git-annex will say: "As far as I know, you have copies of that file on the backup server, and on the Seagate drive." I think the author wrote it so that he could go offline in the mountains of Japan and work without the internet? There might have been hot springs involved, if I recall the story correctly. Anyway, git-annex is a ridiculously geeky tool, and I do not recommend it to anyone but the most esoteric of power users.

Anyway, modern versions of git-annex do not work on the Synology. And the Synology had a security update that prevents me from using rsync, unless I pass "--rsync-path=/bin/rsync". So anyway, I'm going to get all my Spanish study materials back, but it's going to require drastic measures and another 10 hours of data copying. This time I bought a tiny portable Samsung T9 SSD drive with 1 TB of storage, which I'll use to store the big files.

Deciding what to build first. In order to build some tools, I need to focus on getting something useful running. Like, the tiniest possible thing. I have two choices:

1. Build a video player with bilingual subtitle support, and a scrolling list of subtitles. I've done this 2 or 3 times before.
2. Build an incredibly basic Anki knockoff. This is fun and not too hard, but it has more moving parts.

I ultimately want to do both. Anki is good, and AnkiDroid is amazing. But if you're building any kind of external card creator tool, interfacing with Anki is incredibly cumbersome. The card-creating process needs to be seamless, without any export/import steps. Also, like I mentioned, Anki's defaults are all wrong for people who can create 500 cards at a click. So rather than encouraging students to read a whole essay on "how not to torture yourself with Anki", why not fix the UI to encourage good habits?

I should probably start by building the video player, because it's tiny (assuming I use HTML video, which I know far too well). But I already have the SRS sketched out. But either way, my goal needs to be to get native media into the UI ASAP. In the meantime, I can keep watching Avatar, and eventually switch to looping the dialog in the background. Substudy has a cool tool which extracts just the dialog as an MP3 playlist, and leaves out all the long boring sections with no Spanish.

[jeffers]

What is substudy, and how geeky do you have to be to use it?

I'd love to be able to extract just dialogue sections from some of my old DVDs, to listen to like podcasts.

[emk]

What is substudy, and how geeky do you have to be to use it?

Substudy is a tool that I wrote. It's a lot like Subs2SRS, but you need to run it in a terminal. Right now, it's a little bit of a beast to get working, but I'll release a new version with pre-compiled binaries for Linux and Mac shortly.

To turn a video into MP3 dialog tracks, you would open a Terminal and run:

Code: Select all

substudy export tracks myvideo.mp4 myvideo.srt


...where the SRT is the subtitle file. This will output a directory of MP3 files plus a playlist.

There are some great tools for converting and time-shifting subs, including Subtitle Edit and several Mac tools. Once you have an MP3 and an SRT, substudy will do everything else for you. It might be a command-line tool, but it automatically converts encodings, detects major languages, etc. And in keeping with my personal philosophy, it doesn't have lots of options—it tries to have really good defaults baked in. But I'd recommend waiting until my next official release, at least.

Currently listening in the background: "No Hay Nadie Como Tú". This is a fantastic beginner listening exercise—fast but well enunciated, with lots of basic vocabulary and parallel constructions.

(Warning: the music video is Not (entirely) SFW.)


They seem to have some other great stuff like this, too!

[MorkTheFiddle]

What is substudy, and how geeky do you have to be to use it?

I'd love to be able to extract just dialogue sections from some of my old DVDs, to listen to like podcasts.

You could try extracting the audio from the DVD using VLC Media Player and then deleting out the gaps in the conversation with Audacity.
VLC Media Player and Audacity are both freeware fairly straightforward to use.
There are instructions
here for using VLC to rip audio.
Here is a Youtube video with instructions about removing gaps.
NOTA BENE!!! I do not have access to a computer with a DVD drive, so I cannot test this out myself. But in the past I have used Audacity to remove gaps.
Finally, waiting for emk to finish his project might be the best idea. I my self am looking forward to it, because his Substudy effort is a work of genius.
Rdearman has a 5-part series of videos about Substudy beginning with this:
It is in truth a fairly involved procedure.

[jeffers]

What is substudy, and how geeky do you have to be to use it?

I'd love to be able to extract just dialogue sections from some of my old DVDs, to listen to like podcasts.

You could try extracting the audio from the DVD using VLC Media Player and then deleting out the gaps in the conversation with Audacity.
VLC Media Player and Audacity are both freeware fairly straightforward to use.

This method would leave in action sequences, right? (Due to sound effects and music). I was thinking what emk does somehow extracts audio based on the timings of the subtitles.
However, once using Audacity, I suppose it might not be too hard to delete the non speaking scenes. I might have to try this.... now to find a DVD drive somewhere...