Solving the TV problem, part 2

AKA, Subs2srs

In the last post I talked about a way of cutting up podcast episodes or other target language audio, turning them into Anki flashcards, and efficiently transferring the audio into your brain.

This time I want to talk about a way of automating the same process and applying it to TV shows or movies.

Unsurprisingly, I’m not the first person to come up with this idea. Someone wrote a program just for turning video files and subtitles into Anki flashcards. It’s called Subs2srs.

I won’t spend much time explaining how it works, because the sourceforge site does a good job of that already. I’ll just talk about my experience using it.

[Note: My only complaint about Subs2srs is that it only works for Windows. For a while I tried setting up a Windows emulator on my Mac, but this turned out to be too much trouble. This kept me from trying Subs2srs for a while, until a friend graciously lent me her PC.]

Finding material

The hard part with Subs2srs is just finding the files to work with. You need to find a video file and a matching subtitles file in the same language. I haven’t found any strategies that I’m confident enough in to recommend, and in general this can take some trial and error and some adjusting depending on your comfort with downloading things.

Perhaps the safest method for getting a video file would be borrowing a DVD from the library and ripping it onto your computer with a program like Handbrake.

For subtitles, there are several websites that share free subtitle files in various languages. It may take more or less time to find this, depending on the specific movie or show, and depending on the language. There’s a site called Kitsunekko that’s dedicated to subtitles for shows (and some movies) in Japanese, English, Chinese, and Korean.

So far I’ve only tried Subs2srs with Japanese, as resources are relatively plentiful. I’ve tried various anime as well as the TV show 深夜食堂 (Midnight Diner). After some time, which depends on my number of new cards per day and the length of the show, I can watch the original show/movie with pretty much total comprehension.

There are tradeoffs when it comes to choosing the content. In general, it’s good to choose content that you’re interested in, but some consideration should also be given to usability. I enjoyed doing the reps for Fullmetal Alchemist and Attack on Titan, but most of the language in those shows I have rarely had the chance to use in daily life. Midnight Diner is a little better, though some accents are hard to understand, and there’s a lot of slang that I still don’t have enough experience or context to get the hang of. This probably just means I need more practice or some exposure to this kind of speech in real conversation.

Shortcut: use a pre-made deck

Maybe you don’t have access to a Windows PC. Maybe you can’t be bothered to make your own Subs2srs deck. Or maybe you just don’t know where to start. Luckily, some deck makers have been thoughtful enough to share what they’ve made. The biggest repository of Subs2srs decks I’ve found is It’s unfortunate that it’s limited to Japanese, but if you’re learning Japanese, there’s a lot here to start with. Anki Web sometimes has some subs2srs decks as well, but these tend to get taken down.

Another option is to find someone who has already created their own decks and reach out to them (I know at least one such person…). They may be willing to share their decks with you. It stands to reason that someone who is passionate and nerdy enough about language learning to make their own decks would also be excited to find someone else who might benefit from their work.

Case study: Tampopo (1985) – extracting subtitles from a .mkv

One movie I fell in love with recently is Tampopo, a “Ramen Western” bizarre comedy about two truck drivers, a single mother, and a motley gang of other characters on a quest to make the perfect bowl of ramen. I don’t know how everyday or usable the language is in this movie either, but I decided that I wanted to learn the lines anyway, just for fun.

The only hitch was I couldn’t find the subtitles file anywhere. The movie itself was an .mkv, which came with hard-coded subtitles. But Subs2srs needs a separate subtitles file to make the cards.

This is where another program came in handy, the descriptively titled MKVExtractGUI-2, also Windows-specific. I was able to use this to get a separate subtitles file out of the Tampopo .mkv. I believe this uses optical character recognition. As a result, the subtitles are tiny images instead of text. This comes with the downside that I can’t copy and paste the subtitles into a dictionary or Google Translate when there’s something I don’t understand. But it’s not a big deal: Subs2srs also lets you add a native language subtitles file, which can go on the back of the card alongside the target language subtitles.

Here’s an example of what one of my cards looks like. For the front of the card I see the still image an hear the audio, and for the back I see the subtitle.

I’ve only been reviewing these cards for about a week, at only five cards a day (I’m also doing Midnight Diner, and trying not to get overwhelmed). I’ll add more news here once I’ve made more progress.

More languages

As I said, I’ve only tried Subs2srs so far with Japanese. I’m curious to see what it’s like in other languages, like German, Chinese, and Korean, and someday maybe even Thai or Taiwanese Hokkien. I’m also interested in starting to compile a list of movies or shows particularly suited to Subs2srs, or for which it’s easy to find video and subtitle files. If you do give Subs2srs a try, I would love to hear about your experience.

Update June 9, 2017: I’ve been doing the Midnight Diner and Tampopo decks for a couple weeks now. I also created a German deck from the TV show Deutschland ’83 a couple days ago. Here are my impressions:

Midnight Diner – There’s quite a lot of slang and domain-specific vocabulary, e.g. stuff related to the occupation of the main characters of the episode. Most characters speak pretty fast, and some seem to have strong accents, but the subtitles are pretty precise and the audio quality is clear, which makes up for this.

Tampopo – There’s an even more diverse array of accents, slang, and silly, exaggeratedly pompous language (like when the “ramen master” is explaining how to properly show respect for the different ramen components). The audio quality isn’t always very clear, and the subtitles sometimes elide parts of the speech. In some cases this makes it impossible for me to really learn the line. I think this is fine, though. I can toss the cards that are too much of a pain to use. The theatricality of the speech has made it fun to learn so far.

Deutschland ’83 – This is a recent thriller about an East German youth who gets blackmailed by his aunt into working as a spy in West Germany during a nuclear crisis. As such the language is a little bit advanced, but the entertainment factor helps make up for the difficulty. It’s been tough to find German subtitles for German movies and TV. I found one website with lots of German subtitles, but I haven’t yet figured out if it has subs for German shows or if it’s all aimed at foreign-language material.


Nihongo Shark is another site that has a thorough discussion of Subs2srs and how to use it to learn anime (interesting that the majority of Subs2srs users seem to be focused on Japanese. Then again, Anki itself, which isn’t tied to any particular language, gets its name from Japanese (暗記). Coincidence?). The creator at one point uploaded some of his own anime decks to Anki Web, but they’ve since disappeared.

Learn Any Language has a page that explains how to use Subs2srs, and contains some links to other users who have tried it with various language.

Solving the TV problem, part 1

Two months is the longest time I’ve gone without posting since I started this blog last August. This is a major milestone, folks.

Not a bad time to take a step back and recap what I’ve done so far. Here’s a timeline:

Aug '16 - German, 4 hours a day. 
          Learn about language acquisition theory.

Sep '16 - French, < 4 hours a day. 
          [current languages: German, French]

Oct '16 - Travel to France. 
          [German, French]

Nov '16 - Korean. 
          [German, French, Korean]

Dec '16 - Travel to US. Stop actively learning French.
          [German, Korean]

Jan '17 - Indonesian 12-hour challenge. 
          [German, Korean]

Feb '17 - Persian lite. 
          [German, Korean, Persian]

Mar '17 - Event planning 101. 
          [German, Korean]

Apr '17 - Taiwanese Hokkien. Revive Japanese. Stop Korean.
          [German, Japanese, Taiwanese Hokkien]

May '17 - Thai (Terry Hsieh's class). 
          [German, Japanese, Taiwanese, Thai]

There’s a lot to record about my adventures with Taiwanese and Thai, but right now I want to talk about something else.

Space repetition systems + TV programmes


The problem

The best way to learn a language is to surround yourself and constantly interact with people who speak it. That’s how Terry Hsieh learned Quechua in two months (without taking any classes). If you can’t achieve this level of immersion, having a few good friends or a significant other who speaks the language with you on a daily basis is a close second.

If this doesn’t work out either, the next best thing is inundating your eyes and ears with movies, TV, videos, radio — and, yes, other people speaking to you — in the target language. Using this technique, Khatzumoto from AJATT, for instance, got fluent in Japanese in about 18 months.

In this day ‘n age, anyone with a good enough internet connection has this method available to her, in theory. But this is where it gets tricky. With so many options out there, what should I watch? And how? And how much? 

Such a plethora of choices can easily lead to paralysis. Furthermore, the field is fraught with danger. What if I turn on the TV or the YouTube to a German channel, but then I get bored after five or so minutes because my German isn’t good enough and I have no idea what they’re talking about? Or what if I do find something that I like, but there are still parts I don’t understand and I miss important plot points?

This has become a big problem for me. I find a TV series or movie that I really like, and then I have to choose between:

  1. Watching it without subtitles. Soon I’m Googling the plot (in English) or turning on subtitles to find out what I missed.
  2. Watching with subtitles once, then watching a second (and third, fourth, and fifth) time without. (These latter times never actually happen.)
  3. Finding a series I care little enough about that it doesn’t bother me to miss a lot. But then, what’s the point? If it’s a TV show, I usually don’t make it past two or three episodes.
  4. Watching something I’ve seen before in English, dubbed into my target language. Again, there’s almost no point. I rarely watch movies a second time, and usually get bored when I do.

I just can’t win. There is seemingly no right way for me to learn languages with movies and TV.

DIY pilot program for a V-212 helicopter

What I really need is a pilot program for a V-212 helicopter. Except, you know, replacing the helicopter flying part with “Japanese comprehension for a specific movie or TV show.” I plug it into my brain, and then when I watch the movie I miraculously understand everything!

That’s impossible, right?

Nope. Totally doable. The key is a combination of an SRS like Anki, and a way to chop up the TV show, movie, or other media into bite-sized pieces your mind can absorb. Sure, it might take a few weeks instead of a few seconds, but the result is just as rewarding.

Making the cards

This is what I’ve been doing with the podcast Der Explikator, for instance. I take a story that seems interesting, but that I can understand only about 20-30% of on a first listen. I download the mp3, open it with Audacity (any audio editing software will do), and start the card creation process, which looks like this:

1. In Anki, add a new basic card. 
2. Select a bite-size portion of audio, usually around 6-7 seconds.
3. Export the audio portion as an mp3 and drag-and-drop into the 'front' field of the Anki card.
4. Select the corresponding text in the transcript on, copy it into the 'back' field of the Anki card.
5. If there are any parts of the transcript I can't understand, use Google Translate to get the meaning, and maybe copy a few definitions onto the back of the card as well. 
6. Go back to step 1 until the whole podcast has been made into cards.

It took me about a minute per card when I first started, but now that I’m used to it I think it’s closer to 20-30 seconds per card. I’ve realized that this card creation process isn’t wasted, either, because it saves me from having to look up definitions later.

Using the cards

So I’ve made all my cards. Now what? Now I just start reviewing them with the Anki app on my phone. 10 new cards a day is more than manageable — depending on the difficulty, it might take 10-20 minutes a day to get through 10 new cards plus whatever reviews of older cards come due. At this rate, if the average length of audio is 6 seconds, I can do 7 minutes of material a week. A typical podcast is about 15 minutes, so in about two weeks I’ve encountered all the cards, and mastered most of them.

What does an Anki ‘rep’ actually look like with these cards? I listen to the audio and shadow it, and try to understand it. I repeat the audio several times if need be, without looking at the transcript. If I can accurately shadow and understand the whole thing without looking at the transcript, then I count that as a correct answer. If I can’t, I practice a few times while looking at the transcript, and then mark the answer incorrect.

The payoff

A couple weeks have passed. I’ve seen all my cards, and reviewed them until I can not only understand but pretty accurately shadow them all. Ok, now what was the point of all that?

This is where I listen to the complete podcast again. Remember how I only understood 20-30% last time? Oh wait, this time I can understand all of it! Not only that, I can shadow right along with it at normal speed with no difficulty. And because I learned it all in context, it’s easy to recall and use any of the phrases in conversation when I need them.

Was it worth the time I put into it? Let’s do the math. This is after 15 days at about 15 minutes at day, so a little under four hours total. How much time would it take to get to this level of familiarity with 15 minutes of foreign language audio using other methods? I’m not sure, actually, because I’ve never gotten this familiar with such a long segment of non-English audio using other methods. I usually get too bored or give up before it happens.

I can also say that the Anki reps are pretty painless. Thanks to the slight gamification of trying to answer cards correctly, the reps are even somewhat fun.

The next level

For a long time I’ve only applied this method to a few podcasts, and no TV shows or movies. Many podcasts don’t publish transcripts. It takes time to extract the audio from a TV show or movie, find a subtitle file if there is one, and even more time making the cards. Compared to podcasts, a lot of shows aren’t as nice to listen to without any visuals. Plus, hand-making every card does take time.

Well, it turns out there’s an easier way. In the next post I’ll talk about Subs2srs, a program that automates the card creation process.