An Entirely Superficial Comparative SRS Review

I’ve talked an awful lot about using Anki on this site, but to date I haven’t said why (or even if) it’s better than the competition.  In my opinion, it’s the things each program doesn’t do which makes each of them inferior to Anki, and so that’s going to be the focus of this review.

Supermemo

Supermemo has an interesting feature it calls incremental reading, wherein you can easily pick out information and sentences from websites (without lots of tedious copying and pasting) and I prefer its image handling and default screen layout to Anki’s.  The dealer-breaker for Supermemo is the OSs it doesn’t run on.  There’s no native support for Linux or Macs, and even getting it running on Vista (or presumably Windows 7) is difficult.  The retention algorithm has also come under fire for being over-complicated to the point where it no longer works, although I haven’t used the program for long enough to test this.  The program also costs money; the most recent incarnation is a steep $45.

Mnemosyne

Mnemosyne beats Supermemo in both the OS and price stakes – it runs on Windows, Linux and Macs quite happily, and is free to download and use.  It also has a useful how-to-use it shows you the very first time you run the program, which explains all its features – this is something Anki lacks and could do with.  Personally I dislike the tiny window it appears in when you start it up each time, the lack of graphs and the weak card browser – check out some screenshots here to see what it looks like.  It’s otherwise a reasonably good alternative to Anki since it includes sound, image and LaTeX support, although purposefully hiding away configuration options in a text file loses it points.

Anki

Free.  Cross-platform.  Powerful options for configuring and manipulating cards.  I’d point to the single best feature of Anki as its fact model – this lets you create as many card-types as you like from a single set of information (for example, to train listening, reading and writing with minimum time spent inputting information).  To the best of my knowledge, this is not provided by the other programs.  Anki has an extensive selections of pre-made decks and plugins to suit most tastes.  Other plus-points go to the great community and super-responsive author (who incidentally does feature requests, if there’s a feature you want and you’d like to support him).

Final Word

So there you have it – that’s broadly why I’d recommend Anki over the other two programs.  I actually tried Mnemosyne first but felt it was too basic; I trialled a version of Supermemo and disliked its over-the-top clunkiness.  Anki sits in the Goldilocks region of simultaneously having functionality and easy-of-use.

Things I’d like to see for the future (for all three)?  Prettiness.  No, really.  Canonical (who release Ubuntu, a superior alternative to MS Windows) have twigged recently that something open-sourcerers need to get right is appearance.  A subtly textured background for reviews, for example, would make any of the above applications it look super-swish.  Rounded edges and more colourful icons would be a treat too.  Little things, y’know?

Working With Premade SRS Decks and Textbooks

I recently made available the Yip and Matthews Anki decks I’ve been working on for the last few months. (If you want a copy, send me a photo showing the book(s) together with your email address written on a sheet of paper next to them; bear in mind that although it already includes 2500 cards it’s still a work in progress).

Someone asked though how useful it was to use other people’s decks. Here’s the answer: it depends on what the source material is. If all the sentences come from TV shows and books and whatnot – an “organic” deck, tailored to the creator – then it’ll be a waste of time for anyone else to try using it, because of the ordering of the sentences. It’s hard to see underlying patterns (of grammar/word usage) if all the examples are split up from one another.

If the deck is based around a grammar or textbook, then it’s worthwhile using someone else’s deck – after all, they’ve done all the hard work in inputting sentences. It’s easy to familiarise yourself with new sentence structures and the like because similar sentence types are grouped together (as you’d expect from a textbook).

So, what’s the best way of using a textbook and its accompanying SRS deck? I can only speak from experience, but I started out by working through Basic Cantonese, cover to cover – this gave me a good idea of how Cantonese as a language generally worked. This knowledge was enough to start to be able to make sense of dramas and songs. More recently I’ve been using the Intermediate book sporadically whenever I’ve come across a recurring pattern I couldn’t make sense of with my current knowledge (notably 都 and 到) – I’ve been working through the relevant chapters (in Anki) and then getting back to working on listening. Rinse and repeat.

This might lead to a slight problem for people using the deck I’ve put together though since the Intermediate book chapters are out of order. My suggestion would be to suspend all the cards in the deck when you first get it and unsuspend the relevant cards as you work through each chapter. Each card is tagged by chapter name so it should be easy enough to find what you’re looking for – the Anki browser has a decent filter system which can work by tag, so it should be even easier.

In other news: Eric (of When English Attacks) has forked his blog – he’s now also running Attack! Language which is a dedicated how-to-use for SRSs. (In other words, competition for me. ) It’ll be interesting to see how it develops!

Yet More on Anki – Ultimate Cards

I’ve written quite a few posts on using Anki now – it’s an integral part of my language study and I hope that if you haven’t already tried it you’ll give it a go.  For me, my Cantonese decks are a huge archive of knowledge accrued thus far.  I think of it as scaling a mountain, sinking each pick into the cliff face a few inches above the next to make sure I can’t fall.

Anyway, today I’d like to share the most recent incarnations of my main Cantonese deck.  I think (after much experimentation) it’s really approaching a point where it’d be hard to improve on it further.

Comprehension Cards

Quite apart from the typographical perfection (isn’t the balance of font sizes and line spacings to die for? ) I think this encapsulates everything you might need as a Cantonese student.

The Jyutping in the question makes it clear where words start and end (not necessarily apparent from the unspaced characters) and also allows tones to be drilled into the learner’s head.  If the cards had audio, they’d be on the question side.  The sentences are so far predominantly from Yip and Matthews.

The answer side includes a translation (so you know exactly what the Canto means), characters (see section at end) notes on new words and grammar and/or brief definitions, and the card’s tag (decoration).

Active Recall Cards

Cloze deletion may seem at first to be a slightly perverse way of practicing Cantonese, but I promise it’s not!  The two above card types were actually developed through tinkering with Japanese decks – I wanted a time-cheap way of maintaining my knowledge of Japanese and so three months ago I downloaded the Tae Kim deck.  I was surprised that the format was that of Cloze deletion cards (broadly similar to the active recall card above) – clearly the idea was to test active recall, but it wasn’t really something I was up to, having not really done much Japanese for a year.

To get around this, I rejigged the format so that it simply tested passive reading/comprehension ability (similar to the card in the first screenshot).  All I had to do was read and understand kana – this was sufficiently easy to do, so I stuck with the format for a month and a half.

I then decided I’d try the Cloze deletion cards again as an experiment – suddenly, the cards I’d originally struggled with were a comparative breeze.  They were forcing me to actively recall bits of Japanese grammar – and it was easier because I’d already seen each sentence six or seven times before.

So, to cut this long story short, I came up with the two-tier system above – learn a sentence and its meaning, and then remove select parts of it to improve active recall further down the line.  Using Cloze deletion means you can test active recall of specific words – this works better than reviewing words individually (i.e. vocabulary lists) because the extra context developed through comprehension review helps aid accurate recall.

Characters

Although I wouldn’t recommend fresh leaners of Cantonese bother learning (or even looking at) Hong Kong characters, I’ve included them in some of my cards for several reasons:

1. I’ve already invested enough time in them that it’d be a waste not to continue
2. It’s good typing practice
3. I have to look up words in CantoDict anyway; it doesn’t take much effort to copy characters across into cards
4. If I’ve copied out subtitles from a drama, it often doesn’t take much extra work to alter them for the Cantonese
5. I might be able to use them in future cards
6. I’m good at wasting time on non-essential things

Importantly though, I don’t include the characters on the question side – I tried it for a couple of days and found they were either distracting (alongside the Jyutping) or too difficult.  Completing work on Chinese characters is a project scheduled for five months time, and I need to keep telling myself that the day will come when I can learn them properly.  Divide and conquer. Divide and conquer.

Music – Thoughts on SRSing

Something I’ve been wanting to try for a long time, but never got round to, was trying to SRS music to play on an instrument – as opposed to SRSing songs to sing, which works just fine.  I still haven’t got round to it, but here are some ideas…

SRSing

SRSs are useful because they allow us to revise information we already know far more efficiently than if we were left to the scheduling.  They are best suited to small bits of information – for example, names of capital cities, or example sentences of a language we’re learning.

The main limitation of SRSing for languages though is that, by definition, you can only look at fragments of conversations or text.  It is a skill in itself to follow through an entire TV show or book – or even just a conversation or passage.

Music therefore poses something of a problem – typically, we want to learn complete pieces, which unaltered are unsuited to an SRS.  Even scales take up to 30 seconds to go through (depending on how many times you’re practicing it and how many octaves you’re running through) – this is longer than the ideal time for an SRS card.

The question then is, can we break up pieces of music such that we can productively use an SRS to review them?  I’d say tentatively… yes, although questions need to be asked of whether a scheduling algorithm designed for thought-memory is transferable to muscle memory.  As far as I’m aware, this has not yet been tested.

Possibilities

Good music teachers will tell their students to work on difficult bars of music in isolation.  Once learnt, an individual bar should lend itself well to SRSing, since it’s short.  Building up complex phrases of music progressively should optimise the student’s ability to maintain their ability in what they’ve learnt.

Scales should be quite easy to SRS too – a card could just say “F# Major, three octaves, 240bpm” for example.  I used to (and probably still should) have a deck of cards with each scale written on it – each practice session I’d shuffle them and run through all major and minor scales one by one.  An SRS could easily be adapted to allow this.  Guitar chords could work in a similar way.  You wouldn’t be creating question-answer pairs either – the “question” would be to play a piece of music – this is an opportunity for redesigning the SRS specifically for musicians.

Clearly it would not be so effective for professional musicians who practice for 40+ hours per week – their livelihood depends on them being able to accurately assess what they need to practice, and so they’re good at it.  It would probably be most appropriate for people who practice for a few hours each week (like me), who aren’t so good at identifying the most important things to go through each session.

Limitations

One problem would be in determining what each card should test – after all, the most effective SRS cards only test one thing at a time (e.g. pronunciation/vocab recognition/grammar comprehension etc.).  You’d need to note on each card “tone” or “tempo” to know what was the focus of the card – clearly this would take extra time in creating the cards.

Another problem would be knowing where to position the machine displaying the SRS.  You’d have to put a laptop or iPhone on top of your piano, or balance it somewhere high up enough to be visible to a standing flautist or saxophonist – potentially dangerous for the electronics.  You’d further have to keep tapping a key or screen to tell the SRS when to move on.

The biggest difficulty though would be getting sheet music into Anki in the first place.  This would require some new programming.  Anki is integrated with LaTeX, and so it could also be integrated with Lilypond (open sauce notation software that works like TeX – it compiles sheet music from a text-based file), so that the user could easily and quickly create cards.  I’ve tried importing .jpgs from scans of sheet music into Anki, but it’s a prohibitively long process and produces slightly unattractive results.

To overcome the problem of marrying computers and instruments, you could perhaps have the SRS compile a sheet of things to practice each day, which you could print out.  You tell it after your complete practice whether or not you finished everything to your own satisfaction, noting any sections that were particularly difficult (equivalent to marking a card as “hard”).

Conclusion

If the process of adding cards could be reduced to less 30 seconds per card, it might be worthwhile to SRS music for instrumentalists – otherwise though, the time optimisation of practicing would be offset by the initial time investment in setting up the SRS in the first place.  Also, the scheduling algorithms would probably need revision – this is probably the work of several PhDs.

If anyone’s tried SRSing music already, do share – if not, 1) have you considered it, and 2) how would you try it?

- E

Progressive Grammar

Following on from the last post on why grammar is useful, one thing that’s really helped me crack certain elements of Japanese (and subsequently Cantonese) sentence structure is a little something I like to call “Progressive Grammar”.

It’s an intimidating name for what’s fundamentally a very simple thing. Take a sentence (that you might want to SRS) and rip out all the adjectives, adverbs and other non-essential words and start with just that – that’s one Anki card.^[1]

As an English example, we might start with “That afternoon, the kind professor gave her a beautiful bouquet of flowers”; this can be stripped down to “The professor gave her a bouquet of flowers” without losing the fundamental meaning of the sentence. The adjectives and adverbs are peripheral, and whilst they make for more interesting writing, they’re not necessarily easy to digest for a learner.

What we want to do next is to add back in the sentence each word, one by one, and create a new SRS card based on each:

“The professor gave her a bouquet of flowers”
“The professor gave her a beautiful bouquet of flowers”
“The kind professor gave her a bouquet of flowers”
“That afternoon, the professor gave her a bouquet of flowers”
“That afternoon, the kind professor gave her a beautiful bouquet of flowers”

It’s good practice with SRS cards to try to make each card test you on only one piece of knowledge. By deconstructing sentences in this way, one can gain a much better appreciation of how sentences are typically formed, without being overwhelmed by having three or four comparatively new words/grammar points to contend with. It’s also great for quick repetition of words, which in turn makes leaning new vocabulary easier. Here’s a Cantonese example:

“lei5 bong1 ngo5″ (you helped me)
“hai6 lei5 bong1 ngo5 ge3” (it was you who helped me)
“gam1jat6 lei5 bong1 ngo5″ (today you helped me)
“gam1jat6 hai6 lei5 bong1 ngo5 ge3” (today it was you who helped me)
“kam4jat6 hai6 lei5 bong1 ngo5 ge3” (yesterday it was you who helped me)

Notice you can also sentence hack, by substituting words that are grammatically equivalent (in the last example, a time-adverb). Because you’re practicing the same pattern over and over, the precise meaning can be absorbed quickly – and you don’t even need to trawl through verb tables, since all you’re learning is “this sentence type in L2 corresponds to this sentence type in L1″.

It’s really all just the same ruse that Heisig uses to teach you Chinese characters – break something complicated into its constituent parts, and work up from there – adding a single element at a time and doing so in a logical order.  One logical step at a time is always the quickest way to learn anything.  Promise.

P.S. As a side note, you can also apply a similar sort of thing to tricky musical passages – rather than trying to play an entire difficult phrase in one go, it can be helpful to play just the first note, then just the first two, then just the next three, and so on, until the phrase is under your fingers.  Using some tricks for practising scales wouldn’t hurt either.

^[1] Work on just one clause at a time too. It might be helpful to work with a cap of six words for the most basic version of a phrase.

A New Approach to Cantonese

I know I’ve wittered on ad nauseum about how wonderful and intuitive Chinese characters are – and actually, I still think it’s true.  They can give a reader hints as to the meaning of unfamiliar words in a logical, reasonably consistent way.  I’d still recommend any student of Mandarin to learn the writing of the characters before tackling the rest of the language, as per AJATT’s specification.

Cantonese, on the other hand, is only confused by transcriptions of the vernacular.  It would be fine if there was plenty of literature available written using Cantonese characters, but… there isn’t.  There are only smatterings here and there (although some Wikipedians have been making stirring efforts to improve the situation)  – and given the far greater proliference of Standard Chinese, it can occasionally be easy to mix up the two.

On the other side of the fence, it can certainly be useful to know the writing of certain characters so that it’s obvious when grammatical components recur; sometimes there are tone changes for a given character that make patterns hard or impossible to spot if a student has only Jyutping or Yale to rely on.  Without the characters, it sometimes feels like you’re just stringing together apparently disparate, orphaned sounds.

Of course, Cantonese is not helped by the inexplicable way in which songs are penned using Standard Chinese syntax – to adequately sing along to Hong Kong songs (e.g. in a karaoke lounge), one has to learn how to Standard Chinese syntax works – not to mention the special readings.  Worse still, TV shows and films usually have subtitles in Standard Chinese too, making it harder to confirm what was said using the subs, since the two don’t always correspond directly.

Anyway, all of the above factors make learning Cantonese in general objectively harder than other languages – and I’d been particularly struggling with learning each new word’s meaning, pronunciation and writing all at the same time (since Cantonese characters are extremely rare, there were an awful lot of characters that didn’t feature in Heisig).

The Change

So, recently, I’ve dropped Cantonese characters from my Anki cards – I’ve just been using Jyutping instead, thereby only having to concentrate on the pronunciation of words.  Really, I should have done this before – dividing and conquering was the whole point of using Heisig’s method for learning characters in the first place, and it still makes sense to segragate learning readings and writings for new ones.  (I still have all written Chinese cards using characters, though – I haven’t ditched them!)

This has increased my progress somewhat – since I’m no longer struggling to learn characters that are of dubious long term value, I can learn more words/constructions in any given period of time.  Naturally, I can go back and learn the characters later if I choose – and because I’m using Anki, I won’t forget those I’ve already invested time in learning.

Since I’m not bothering with characters right now, this has enabled me to take every example sentence from my grammar and input it into Anki in a couple of days – before, I was worrying about everything being written out in HK characters, and copying anything out took way too long.  I now have several thousand new examples in my deck – plenty to be getting on with.  This has had a secondary advantage – whenever I come across new words in films or radio shows, I now have a convenient resource of sentences they can be dropped into.  It’s quite difficult to transcribe an entire sentence from sound alone; it’s easy to listen to and look up the one word and deftly insert it into a premade example sentence or six.  This then reinforces grammar points as an added bonus.  (Incidentally, this is just an elaboration of a great idea for sentence hacking Eric of When English Attacks had – I can’t claim credit!)

I’m still learning some characters as I come across them, so finishing Heisig was by no means a waste of time – but only Standard Chinese characters.  I do want to be able to read Chinese books, and sing along to songs!  The only downside to ignoring the HK character set from now on is that occasionally I might miss patterns that would make remembering certain words easier – but I think, on balance, that it’s a price worth paying.