what3words - mapping the world using three-word phrases
Posted: Tue Jul 07, 2015 9:22 pm
Recently I discovered this: http://what3words.com/
No, it's not a new word game. It's an interesting way of mapping every 3m square on the earth to an easily memorisable three-word address. For example, Nelson's Column is at beans.again.voting. The entrance to the COLIN Travelodge is perky.copies.sprinter, and the entrance to the Memorial Hall is superhero.clattered.pile. Incidentally, the company was co-founded by Jack Waley-Cohen, famous for setting questions on Only Connect, encouraging a bunch of us to apply for OC earlier this year, and almost but not quite turning up to superhero.clattered.pile in the process.
Anyone who has had to write down a grid reference or a latitude and longitude, or bemoaned that their satnav has taken them to the "wrong" part of the destination postcode, should appreciate how useful this addressing system can be.
The algorithm assigns a three-word address to all 57 trillion 3-metre squares on earth, and that already makes it better than postcodes in that these addresses are easier to remember, more precise, cover the entire world rather than just one country, and lookups of them aren't jealously guarded and rationed by Royal Mail.
A few other features are quite curious, though, and IMO they make this just a neat and interesting idea rather than an amazing and world-changing one.
The algorithm they use is not public, so every time you want to translate to and from a three-word address, you have to use their API, or their application, or their website, so it's not really an open system. You couldn't make your own implementation of the algorithm and use it freely, and if you're using the addressing system commercially then it isn't free to use. At least their "free for individuals" policy means it's a bit more forthcoming than Royal Mail have been with the postcode database - there wasn't even a complete public list of postcodes until 2010 when Ordnance Survey released CodePoint Open, which gives the latitude and longitude of the centre point of each postcode.
There's no effort to make nearby points share similar addresses - in fact, quite the opposite. Similar-sounding addresses are deliberately assigned to points huge distances apart. This has the obvious benefit that if you mistype or mishear an address and get it wrong, it will almost certainly be very obvious that it's wrong. For example, suppose someone told you that the best place to stay during COLIN weekend is perky.copies.sprinter. If you misheard it as perky.copy.sprinter you'd find that was in Alaska, and if you misheard it as perky.copies.printer you'd get a point somewhere in Morocco, but you definitely won't get a plausible-but-wrong Lincoln address.
There is a vague pattern to the word assignments which is beneficial - populous areas get addresses consisting of short or common words, and unpopulated areas such as the middle of the ocean get the longer or less well-known words.
However, personally I think the apparently random addressing is a bit of a flaw. Even postcodes, which so far have been beaten hands down by what3words on every criterion except being well known to the public, have the property that by looking at the first one or two letters of the postcode you can see which part of the country it's in, and if you live in that area you'll probably have a fair idea which district is represented by the number(s) that follow it.
With a postcode, latitude/longitude, grid reference, or practically any other addressing system, you can normally visualise without the aid of electronics roughly where it is, or whether it's anywhere near you, or whether it's in the same part of the world/country as another address of the same format, or roughly what direction and order-of-magnitude distance you need to travel if you know the address of your location and that of your destination, but none of that is true for what3words. If you're given a three-word address, unless you happen to know that exact address - "oh yes, I know perky.copies.sprinter, I stayed there last year" - you have to get your phone out and look it up to have any idea where it is.
"perky.copies.sprinter, you say? I stayed at chess.idea.surfaces."
"Oh really? Where was that?"
"It's a Travelodge near a Little Chef southwest of Lincoln."
"Wait, that's where I stayed." *gets phone out* "Oh wait, I see. perky.copies.sprinter is the address of the entrance, chess.idea.surfaces is the address of the reception desk."
The decision to make the addresses appear random rather than hierarchical seems like a conscious decision. Some might suggest they want to maximise use of their service and don't want people to be able to guess approximate address locations for themselves. Maybe that's too cynical, and it's just a well-meaning attempt to prioritise error detection over transparency. Still, the idea to represent addresses using words rather than numbers is a good one, and perhaps if you added another word and made an address four words, you could use a short enough word list (about 3,000 by my rough calculations) that you could filter out any words that sounded the same as or similar to other words, and still only use shortish common words (I'm guessing - haven't really tested this), but at the same time have a hierarchical addressing system where it's obvious how it works.
No, it's not a new word game. It's an interesting way of mapping every 3m square on the earth to an easily memorisable three-word address. For example, Nelson's Column is at beans.again.voting. The entrance to the COLIN Travelodge is perky.copies.sprinter, and the entrance to the Memorial Hall is superhero.clattered.pile. Incidentally, the company was co-founded by Jack Waley-Cohen, famous for setting questions on Only Connect, encouraging a bunch of us to apply for OC earlier this year, and almost but not quite turning up to superhero.clattered.pile in the process.
Anyone who has had to write down a grid reference or a latitude and longitude, or bemoaned that their satnav has taken them to the "wrong" part of the destination postcode, should appreciate how useful this addressing system can be.
The algorithm assigns a three-word address to all 57 trillion 3-metre squares on earth, and that already makes it better than postcodes in that these addresses are easier to remember, more precise, cover the entire world rather than just one country, and lookups of them aren't jealously guarded and rationed by Royal Mail.
A few other features are quite curious, though, and IMO they make this just a neat and interesting idea rather than an amazing and world-changing one.
The algorithm they use is not public, so every time you want to translate to and from a three-word address, you have to use their API, or their application, or their website, so it's not really an open system. You couldn't make your own implementation of the algorithm and use it freely, and if you're using the addressing system commercially then it isn't free to use. At least their "free for individuals" policy means it's a bit more forthcoming than Royal Mail have been with the postcode database - there wasn't even a complete public list of postcodes until 2010 when Ordnance Survey released CodePoint Open, which gives the latitude and longitude of the centre point of each postcode.
There's no effort to make nearby points share similar addresses - in fact, quite the opposite. Similar-sounding addresses are deliberately assigned to points huge distances apart. This has the obvious benefit that if you mistype or mishear an address and get it wrong, it will almost certainly be very obvious that it's wrong. For example, suppose someone told you that the best place to stay during COLIN weekend is perky.copies.sprinter. If you misheard it as perky.copy.sprinter you'd find that was in Alaska, and if you misheard it as perky.copies.printer you'd get a point somewhere in Morocco, but you definitely won't get a plausible-but-wrong Lincoln address.
There is a vague pattern to the word assignments which is beneficial - populous areas get addresses consisting of short or common words, and unpopulated areas such as the middle of the ocean get the longer or less well-known words.
However, personally I think the apparently random addressing is a bit of a flaw. Even postcodes, which so far have been beaten hands down by what3words on every criterion except being well known to the public, have the property that by looking at the first one or two letters of the postcode you can see which part of the country it's in, and if you live in that area you'll probably have a fair idea which district is represented by the number(s) that follow it.
With a postcode, latitude/longitude, grid reference, or practically any other addressing system, you can normally visualise without the aid of electronics roughly where it is, or whether it's anywhere near you, or whether it's in the same part of the world/country as another address of the same format, or roughly what direction and order-of-magnitude distance you need to travel if you know the address of your location and that of your destination, but none of that is true for what3words. If you're given a three-word address, unless you happen to know that exact address - "oh yes, I know perky.copies.sprinter, I stayed there last year" - you have to get your phone out and look it up to have any idea where it is.
"perky.copies.sprinter, you say? I stayed at chess.idea.surfaces."
"Oh really? Where was that?"
"It's a Travelodge near a Little Chef southwest of Lincoln."
"Wait, that's where I stayed." *gets phone out* "Oh wait, I see. perky.copies.sprinter is the address of the entrance, chess.idea.surfaces is the address of the reception desk."
The decision to make the addresses appear random rather than hierarchical seems like a conscious decision. Some might suggest they want to maximise use of their service and don't want people to be able to guess approximate address locations for themselves. Maybe that's too cynical, and it's just a well-meaning attempt to prioritise error detection over transparency. Still, the idea to represent addresses using words rather than numbers is a good one, and perhaps if you added another word and made an address four words, you could use a short enough word list (about 3,000 by my rough calculations) that you could filter out any words that sounded the same as or similar to other words, and still only use shortish common words (I'm guessing - haven't really tested this), but at the same time have a hierarchical addressing system where it's obvious how it works.
