I am now in Hong Kong since three months ago, and while some (few) people are honestly amazed that I speak Cantonese at the level I do now, I’m nowhere near fluency, and have the feeling that I’m progressing quite slowly for the rather great amounts of time I invested. From time to time I try a new approach, to see if it brings me to my goal faster than what I’ve tried before, and some of these times I get lucky.

While being amazingly easy regarding grammar, the usual obstacles for learners imprinted by Indo-European are the tones (I can’t remember them!) and short syllables (makes it hard to apply mnemonics I commonly use in other languages: Albanian falmendirit sticks to my mind since I imagined three (German “drei, zu dritt”) men falling down shouting “Thanks” some years ago), and that it’s not related to the already familiar word pool. When learning languages like French or Portuguese, I very often encounter old friends in new guise. Not so in Cantonese (although Mandarin will be exciting in this respect, should I get around to learning it after Cantonese).

A system to learn Mandarin syllables

I was directed to Serge Gorodish’s blog Country of the Blind by a post by Benny Lewis (Speak from Day 1), who learned to speak fluent Mandarin in three months. While he did not use the mnemonics technique developed by Serge himself, he suggested to give it a try. I have tried mnemonics before, to some success, so I checked it out. Serge’s technique is an elaborate system to remember the sounds of Mandarin syllables, which awed me, and which I could never even have dreamt of developing myself. I tried to adapt it to Cantonese, because after all, they are pretty similar languages, and failed. Serge’s system, very roughly summarized (check his article for the complete explanation), goes like this: Mandarin syllables consist of an initial, a final, and a tone, and sometimes of one sound in the middle. He associates the initial (+ middle) with famous real and fictitious persons (personages), and associates the finals with locations, and uses sublocations for the tone.  This way, he ends up with 51 personages, 12 locations, and 4 sublocations, to encode about 450 different syllables (not counting tones). 苗 miao2 comes up as Marilyn Monroe (mi-)  and the lobby (tone 2) of the Keio Plaza Hotel (-ao). I tried the same approach with Cantonese, but ended up with about 150 items instead of 60 (still less than the about 630 syllables), so I asked him for advice. He very helpfully pointed me in the right direction, so I could develop a system for Cantonese that from my preliminary trials seems to work very well, and which I want to describe here.

Rethinking the system

Looking at the table of Cantonese syllables, it became clear that blindly copying Serge’s approach was unwieldy and inefficient. My idea was to instead use locations for the 20 initials, and personages for the remaining 56 finals. But then I was stuck again, I had no idea how to connect the finals with person names without it all ending in chaos. Serge again helped me out, suggesting that I reverse the finals. This worked perfectly.

The 20 initials are: Ø- (no initial) b- p- m- f- d- t- n- l- g- k- ng- h- gw- kw- w- z- c- s- j-

I use Jyutping romanization, not to be confused with the slightly different Yale system. In Jyutping, J is pronounced like the Y in Yale (within the system, and literally), while in Yale it’s pronounced similar to the J in Jackson. There are other minor differences.

The 56 finals can end in (no final sound) -i -u -m -n -ng -p -t -k, and have one of the eight following sounds before it: -aa- -a- -e- -i- -o- -ö- -u -ü-.

The -ö- and -ü- are my personal notation: -ü- is commonly written -yu- (for example in Jyutping, probably the best Cantonese romanization system), while -ö- is -eo- before -i, -n, -t and -k, and -oe- otherwise. They are written differently in IPA (ɵ vs œː), but as they are clearly distinct by never preceding the same sound, it seemed justifiable to group them both under “ö” (Frankly, I don’t hear the difference, and only a few days ago noticed that both exist :P ). Why “ö” and “ü“? I happen to be a German native, but this also works for Turkish speakers (and Austrians!!) ;)

56 famous people

So considering the eight vowels before the final sound, I created eight categories of personages. Instead of using eight different types of characters, I used four, with males separate from females, like this:

  • -aa- male actors
  • -a-   female actors
  • -e-   male musicians
  • -i-    female musicians
  • -o-   male fictional characters
  • -ö-   female fictional characters
  • -u-   male politicians
  • -ü-   female politicians

Actually, quite some thought (aka trial and error) went into this, because I did not find it that easy to find corresponding personages. One thing is that there are more male than female famous fictional characters and politicians (and my feeling was this is even the case for actors). Serge uses a similar system for Mandarin initials, and recommends to associate the first name of the person with the initial. I mostly chose the last name; I have no idea whether that is worse, but it felt more natural to me. You may notice Hitler on the following list; you see, he’s finally useful for something after all :) Even though I enjoy his follies at Family Guy, I prefer to be on a last name basis with him.

Let’s look at the first batch of personages for how this part of the system works:

-aa-: male actors

  • -aaØ (20): Leonardo (DiCaprio)
  • -aai (20): (Samuel L.) Jackson
  • -aau (13): (Arnold) Schwarzenegger
  • -aam (11): (Bill) Murray
  • -aan (17): (Jack) Nicholson
  • -aang (14): Harpo (Marx)
  • -aap (10): (Al) Pacino
  • -aat (16): (Quentin) Tarantino
  • -aak (17): (Ben) Kingsley

In the likely case that a silent “WTF” crouches up your throat, read on.

Leonardo comes to the fore for every syllable ending in -aa; Bill Murray runs the -aam section, Jack Nicholson is responsible for -aan, and so forth.

You may notice some inconsistencies here (actually, the first three are inconsistent, hmmm…) and in the following groups. Firstly I sometimes used first names instead of last, or did not use the first letter of the name to match the final). I have the feeling that this will not lead to confusion in my system, but your mileage may vary. You may want to create your own list of personages, finding the ones that are most iconic to you. Some of these you may not even know (“Nena” anyone?). If you can make them more consistent this is likely better.

You will also notice that sometimes the initial of the name and the last sound of the final are not the same, but similar. My “regular exceptions” (not sure if that’s a pun) were these: if there’s a š sound (Schwarzenegger, Shakira), use the first consonant after this. L does not occur as a final, so is used for the -Ø (no final sound). H is used for -ng, as it also does not appear at the end. As persons starting with U are surprisingly hard to find, V (hell, U was written like this in Roman times anyway!), W (hey, it’s just two V!!) and F (V can sound like this in German) can replace it. J can be used instead of I, B instead of P and C/G instead of K. Did I mention I had trouble finding enough personages? Again, I haven’t tested this system in depth yet. While from preliminary tests I have the feeling it will work fine, it’s probably safer to have a 1:1 relationship, so if you find enough persons starting with U/I/P/K, go ahead :) Yes, Tarantino is also an actor!

Choose characters as iconic as possible, they will populate your Cantonese universe for some time. For example, my blurry images of Karl Marx and Marc Twain look very similar, so they didn’t find a place in my system. I’ve added some characteristics in parentheses to illustrate how I iconicize them.

-a- : female actors

  • -ai (20): (Angelina) Jolie (as Tombraider)
  • -au (17): (Sigourney) Weaver (in Alien)
  • -am (14): Marilyn Monroe
  • -an (18): (Anna) Nicole (Smith)
  • -ang (17): (Selma) Hayek (in From Dust Till Dawn)
  • -ap (12): (Natalie) Portman (in Leon)
  • -at (17): (Liz) Taylor
  • -ak (9): (Whoopie) Goldberg

-e- : male musicians

  • -eØ (13): (Franz) Liszt
  • -ei (12): (Michael) Jackson
  • -eu (1): (Stevie) Wonder
  • -em (1) (WA) Mozart
  • -eng (12): Heino
  • -ep (1): (Elvis) Presley
  • -ek (10): Cat (Stevens) (as Yussuf Islam)

-i- : female musicians

  • -iØ (7): Lisa Lopes (as Left Eye)
  • -iu (14): (Nelly) Furtado
  • -im (11): Madonna (with cone bra)
  • -in (14): Nena (with balloon)
  • -ing (16): (Whitney) Houston
  • -ip (10): Björk
  • -it (14): Tatu (kissing)
  • -ik (17): Shakira

-o- : male fictional characters

  • -oØ (19): Leatherface (with chainsaw)
  • -oi (13): Yoda
  • -ou (14): Frankenstein
  • -on (4): Neo
  • -ong (19): Hulk
  • -ot (2): Tarzan
  • -ok (19): King Kong

-ö- : female fictional characters

  • Ø (5): Elastigirl
  • -öng (9): Helen (of Troy) (riding a wooden horse; yes, my Greek mythology is a bit rusty)
  • -ök (8): Catwoman
  • -öi (11): Jeannie (with crossed arms)
  • -ön (7): Nike (wearing Nike shoes)
  • -öt (6): Daisy Duck

-u- : male historical politicians

  • -uØ (4): Louis (XIV) (with wig and fat)
  • -ui (7): (Boris) Yeltsin (with vodka bottle)
  • -un (6): Napoleon
  • -ung (17): (Adolf) Hitler (ambassador of Achtung! and Endlösung)
  • -ut (6): (Leon) Trotski (with ice pick in neck)
  • -uk (16): (Helmut) Kohl (alternatively Qadafi)

-ü- : female historical politicians

  • Ø (4): (Rosa) Luxemburg
  • -ün (11): Nofretete
  • -üt (10): (Margaret) Thatcher

I had a hard time coming up with even three iconic female politicians, that’s why this section comes last. I added the number of initials that can be combined with any final in parentheses above. Presley, Mozart and Stevie Wonder will almost never be on stage, as their respective syllable finals -ep, -em and -eu only occur with one initial (gep, lem, deu, respectively), while my universe will be full of the much less iconic Selma Hayek and Angelina Jolie; that’s a pity! I think my group of personages is good enough to work for me, but I’m not entirely happy with all of them; so if you have some more iconic names for the list, that might even fit better with the consistency of the system, leave me a note :)

20 infamous locations

All that’s left to complete the system is to associate locations with the initials. As Serge mentions, this has to be done by you, if you wish to use the system, as you have to be familiar with these locations. Serge uses restaurants, houses of relatives and so on; I use similar locations (former workplaces, homes, homes of friends, places I’ve stayed). I did the same, but for the Ø-initial (for example used when the syllable to remember is aap, so Al Pacino has a place to stay), I use the moon lander (Yes, I’ve been to the moon). I loosely associated these places with the initials where possible (the location for d- also starts with a D), but this is not a must, as pretty soon the association in your mind will be there irrespective such beginner’s crutches. So don’t force it, but use it where possible.

The tones are encoded by sublocations. I chose some that correspond roughly with the tone itself. The first tone is the highest: the sublocation is the roof. The second tone is rising from low to high: on the stairs, in the elevator, on the escalator. The sixth tone is low: outside the building. The fourth tone goes from low to even lower: in the cellar or basement, or going down the stairs leading to it. The third tone is in the middle: the sublocation is anywhere in the building. That is, except for the toilet, which is reserved for the fifth tone.

Mind that this system is missing two syllables, namely “ng” and “m” (the “hm” and “hng” mentioned in the Cantonese syllable table don’t seem to exist). There is no -m in the male politicians section, so Chairman Mao gets to represent “m”; the only section without -ng is female politicians, so Hillary (Clinton) gets the honor to represent Mao’s female counterpart. Both will be restricted to the moon lander; ah well, I’m sure they’ll get along…


I like the title of Serge’s blog, Country of the Blind. It reminds me of the fact that we are all incredibly stupid; after all, we only recently evolved from a species similar to chimps and bonobos. Our brain is essentially a monkey neuro-system on steroids. Try to remember 9 different numbers or names by looking at it for less than a second, then looking away (actually, chimps can do this…). Try to visualize spacetime, or to change your habits to become a fully happy person, realizing that you are one with the world.

The idea behind mnemonics is that while our brain is really bad for storing long lists of boring stuff, it is really good at remembering vivid stories, and constructing web-like connections. This means to use our brain for things that it is actually designed for. Remembering random information (there were 823 pebbles on the way from the river to my home)? No way. A dragon came out of the sky and ate your mother? Hell yeah! Even generations to come will remember this (look at the bible for examples). The great thing is, these stories need not be real, they can be imagined!! (look at the bible for examples). So our task here is to use the building blocks outlined above, and construct stories that are as remarkable, as unusual as possible. The more surreal and colorful, the better. But the most important thing is that you have really imagined this story once.

Consider the following example. is jung6 (J=initial, -ung = final, 6=low tone). It means mostly “to use”. Adolf Hitler to the rescue! In my image, he’s sitting in front of (tone 6) my old university building (initial J), together with the Moon, who is injecting a needle with heroin into Hitler‘s arm. Why the moon? This system can (and should, according to Serge) incorporate the form of the character itself. I learned Cangjie, a typing system that breaks down all Chinese characters into 24 different types of elements. These can then be quickly inputted using a standard QWERTY keyboard. The character 用 is written BQ, or 月手, or “moon hand“. There are many other ways to memorize a character, Hacking Chinese presents some nice methods. In Germany, heroin addicts are sometimes called “User”, that’s why the heroin part effectively gives me the meaning of the character. Hitler gives me the final -ung. I try to imagine this scenario as vividly as possible, with me going by, and somebody asking me, “Who’s that?”, and I, with some resignation, answer, “Ah, that’s Hitler and the Moon, they’re sitting there all day, doing nothing! What can you do?” This example is bizarre enough for me to remember, but this example could be upped even more, for example by imagining the moon injecting microscopic Hitlers into his arm. The general rule is, if you don’t have the feeling that the story at hand has made a lasting impression on you, make it more crazy. Inventing such a story, imagining it, and then letting it go takes only some seconds with some practice. Just trust your mind to bring it back when needed.

Another example: I could never remember the tone of “horse”. The syllable is maa5, so my image is two dead and roasted Leonardos (di Caprio), holding hands, being eaten by a horse on the toilet (tone 5) of my aunt’s house (initial m-). Why two, and why do they hold hands? The character in Cangjie is written 尸手尸火 “dead-body hand dead-body fire“.

Last example: “to carry”, syllable tai4, Cangjie 手日一人 “hand eye one man“. I made this one up on the walk from the bus station to home, and Angelina Jolie carrying bags full of tiny one-eyed men down the stairs to the cellar in the home I grew up in (initial T) stuck. Could be more bizarre, but works.

I’ll apply this system in the near future as much as possible, to get into the habit of using it. If you find any errors, or have ideas for improving it, I’d be very happy to read about it. Thank you Serge for coming up with this!

Update 1: I have now tried the system for several days, and am really impressed with it. Some elements that are not ideal to me are:

  • some personages are not iconic enough. After some days, I forget if it’s Nelly Furtado or Angelina Jolie that act in my mnemonics
  • the roof and cellar are not ideal sublocations, as in some places, I have to make them up (because I’ve never been there, or they don’t even exist), and so after some days I have a hard time to remember in which cellar the story takes place. So maybe better use other sublocations.
  • while it’s not a great drawback, keeping the system as simple and consistent as possible definitely makes it more powerful, because it becomes faster and more effortless. For example, Harpo is a real actor, but also kind of a fantasy figure; A. Jolie appears quite often in my stories as Tombraider, a fantasy figure; I have some problems remembering the personages for -u, because they can begin with W/U/F/V; Monroe has also been a singer; these all work for me, but take an extra second to sort out sometimes, which can lead to unnecessary frustration.
  • my personal favorite at the moment is King Kong on a tree in the middle of my street (L-), with white silk stockings hanging down his ears and laughing wildly. Neatly and instantly gives me 樂 lok6, helpful during Chinese New Year. 新年快樂!!

Update 2: A year or so later :)

  • instead of using roof and cellar, I use: 1: window, 2: inner door, 3: inside, 4: outer door, 5: toilet, 6: outside. This works quite well.
  • I never became fast enough to comfortably remembering new characters “on the fly”. Maybe you need a physicists’ brain for this ;)
  • I am currently “learning” 10 new characters per day, but it became very clear to me that a mnemonic system alone is not enough to plant the complete ecosystem of Chinese Writing into my head. It can be quite helpful to “look at the character once” in its full depth, as advised by Harry Lorayne (I think), and helps to carry the knowledge over to some other task. However, it is very important to then use the hell out of the learned characters, re-incorporating them wherever you can, so you build a true multi-dimensional network of knowledge.
  • I’m not using Cangjie associations anymore (although I use Cangjie itself more than ever :). After a while, more organic sub-forms tend to emerge, so that bigger parts of the characters can be remembered as a whole.