This article is old and needs a rewrite to refer to the newer techniques.
If I had to learn the addresses of a big city so that a customer in my [imaginary] cab could ask me to drive to an address then I would think about using mnemonics to learn the addresses.
Rote memory is the best memory but, if you have an exam coming up and you need to cram for it then it is good to have mnemonics. Let's imagine that I have to quickly learn 100s of street addresses.
Some memory experts might recommend visiting as many of the addresses as you can so that you have a visual memory of them; but to imagine a mnemonic of the road name happening in the street. Since 'Cavendish Avenue' sounds a bit like 'Cave' followed by 'Dish', one could imagine a small cave opening and a huge dish outside it - as if it is a big obstruction in the avenue.
When a passenger asks to be driven to Cavendish Avenue, what should happen is: Cavendish becomes Cave+Dish; and you have a chance of remembering the road(s) of the city where you pictured a cave and dish.
My approach of 'using mnemonic images for the spelling of the start of a word' would not fare well here. If I have imagined the image for CA at Cavendish Avenue or imagined an image for CAV or for CAVE then what happens when the customer asks to be driven to Cavendish Avenue? How can I turn that into CA and think of roads where the CA image was pictured? There would be so so many roads beginning with CA that the technique is very weak; although an image for a 4 letter prompt of CAVE has more hope.
So it looks like the Cave + Dish approach is best. I would say though that it can go wrong too...
Imagine a road is called Affleck Road and you think, "Affleck sounds like 'a flock'; and you imagine a flock of sheep in the road; then a customer one day asks you to drive to Affleck Road. What do you convert Affleck into? A flock of sheep or a flock of sea gulls? Or maybe Affleck reminds you of 'a flick'. The connection can be broken because there is no firm rule about how you will interpret Affleck in future.
That broken connection is one of the reasons why I would not tend to use that approach to memorise foreign language.
Having said that criticism, Cavendish as Cave and Dish was pretty good. So I would not reject the method completely; and, given several iterations of revision, the mental association would become strong.
Maybe there is no super mnemonic solution to the taxi driver question. Far better would be to know from a young age that I want to become a taxi driver; and then to learn loads of roads at my leisure rather than in a hurry.
I wonder if a combination of using a 3 letter image and a 2 letter image imagined in a street would work? Eg. Brooks Street might be thought of as BRO + OK (+ the S which we ignore). The problem with the 3 letter robots system or 3 letter musicians system is that the images are quite samey; they are fine for occasional use but, if 100s of roads are all expressed as robot images or as musicians playing a tune then the samey quality would creep in and make them less memorable, collectively.
If a three letter system is constrained to being just common three letter combinations then it is no longer a 26 x 26 x 26 letter exercise and maybe worth the time to develop - the images would hopefully not be samey. Maybe some of the 1000 male and 1000 women images could be used within it. Eg. a person meaning 'Well' could represent WEL. Any street name that does not fall into that system could be expressed as robot images, etc..
If the AA-ZZ system is made up of sketched cartoon-like objects then they could be imagined as different colours depending on which district of the city the road is in; or maybe colour could mean whether the road is a 'Road', 'Street', 'Hill', 'Avenue', etc..
And what if each AA-ZZ image has its own action? Then a story imagined at a street could be a 3 letter image from the 1000 people lists and that person would do an AA-ZZ action using a coloured AA-ZZ object; and that action could be performed on an AA-ZZ person. I could develop that action system by re-engineering the 1000 actions I already made (see the earlier article).
Another person I could be imagining could be one of the people from the 5000 peg system. So there are many ways to make a visualisation to represent a road name. (When I created the 5000 peg system, I deliberately used syllables that might be helpful in learning a foreign language that is consonant vowel consonant vowel in nature - but then it developed into a peg system rather than a spelling system.)