What3Words Mongolia

Earlier this year I wrote that every 3m*3m place on the face of the planet can now been addressed by just three words:

what3words has identified every one of the 57 trillion 3mx3m squares on the entire planet with just three, easy to remember, words. My office, for example, not my building but my office, is token.oyster.whispering. Tyler’s office just down the hall is barons.huts.sneaky. (Especially easy to remember if you recall this is Tyrone’s office as well.)

Every location on the earth now has a fixed, easily-accessible and memorable address. Unpopulated places have addresses for the first time ever, of course, but now so do heavily populated places like favelas in Brazil where there are no roads or numbered houses. In principle, addressing could be done with latitude and longitude but that’s like trying to direct people to web sites with IP addresses–not good for humans.

The post office of Mongolia has just announced that they will use the system.

Mongolians will be the first to use the system for government mail delivery, but organizations including the United Nations, courier companies, and mapping firms like Navmii already use What3Words’ system.

Mongol Post is switching to the What3Words system because there are too few named streets in its territory. The mail network provides service over 1.5 million square km (580,000 square miles), an area that’s three times the size of Spain, though much of that area is uninhabited. Mongolia is among the world’s most sparsely populated countries, and about a quarter of its population is nomadic, according to the World Bank.

Even in the capital city of Ulaanbataar, not all streets are named. When people don’t have a street address, the current solution is for them to travel to a collection point to pick up their post, says Chris Sheldrick, the co-founder and chief executive of What3Words. People have to write a series of detailed directions, in addition to the address, so that mail-delivery people know where to drop off letters, Sheldrick say.


I've already copyrighted your address so you owe me.

But for the Mongoloian post office to recognize 'barons.huts.sneaky' they will still need to match up the Mongolian three words, in Cyrillic, with the American English system before any mail can be actually transferred.

The system is less universal than it appears, particularly if, as it appears, what3words claims to own all of those 3 word combinations, as noted here - 'what3words will always be free for individuals to use on our own site and apps. If or when we do charge for access to our web API or offline SDK, there will always be ways to use them for free.

In particular, we intend to support fair and equitable use of our core addressing technology. We employ a fee structure that provides qualifying organisations with a range of free and discounted usage plans, in addition to country-based pricing. Qualifying organisations will include humanitarian and not-for-profit entities in any country, and regional and national government and associated organisations registered in countries that fall under the World Bank Low-Income Country (LIC), Lower-Middle-Income Country (LMIC) or Upper-Middle-Income Country (UMIC) categories. Discounts are based on world economic indicator data compiled and published by the World Bank.

Furthermore, we understand that organisations who integrate what3words need assurances about the long-term viability of the technology.

Our goal is for what3words to become a global standard for communicating location. At the moment, the core what3words algorithms and data are not in the public domain. In the future, we may release some or all of our source code – we will continually evaluate the business case for doing this.' http://what3words.com/pricing/

Strangely enough, the world does not need to pay for an address presently - but now, it can.

And the most amusing thing, if this was a technology with a GPL style license, it could actually prove quite useful, without the least concern about any long-term viability of the technology, nor any need to evaluate any business case.

A company has invented a valuable service and you appear angry that they want to make money off this. I don't see a coherent point here.

I doubt the company owns the meta-concept, of translating a location into a mnemonic.

It can't really be that hard to do with a Mongolian dictionary. Use 4 words if you want to make the company feel less bad. A sr project at any local university.

Well then there's no problem, right? If they don't have any value and can be easily duplicated then I am sure Mongolia will do that rather than pay.

One hopes, though perhaps a sales agent is talking to a government official who is not talking to any engineers. It happens.

Assuming anyone actually thinks this valuable, it will be duplicated in much the fashion that many good ideas are - after all, the Internet is pretty much a place where none of the ideas, and little of the software, has a price tag or is proprietary - though some software companies have spent billions before learning that fact.

It is so interesting to see that just because a company could convince any government agency to provide the company revenue for a proprietary solution, we talk about the company's great idea, while ignoring how a government is paying for something that would be cheaper and more reliable in the longer term, using a couple of decent programmers over a few days to work out a similar system - and when the GPL is used, people anywhere can improve the work, without cost.

Generating three word strings from a database to attach to a coordinate grid is not exactly a difficult problem.

It may not be a Manhattan Project level of difficulty problem, but it's not an easy one either, for practical reasons.

Read the link. The system is set up in many languages. They have taken care to avoid ambiguities arising from homophones, and confusion from the use of plurals. The assignment of phrases to places is not random, but seems to have been done methodically.

Sure, anyone can take latitude and longitude and assign phrases to coordinates. Doing a good job is something else.

It is usually a mistake to think that because something is easy in principle it is easy in practice.

So you are saying it might be a week's work, and not just a day. Fine.

But maybe trust programmers when they look at problems as "hard" or "easy."

See anon's comment - a comment that actually seems to understand how trivial this truly is. Companies are always interested in making money, of course, but generally, they need a way to restrict their customer's freedom to do something on their own, particularly in the realm of software. In this case, the word list is the basis to the coordinate system - even though, as decades of human experience using the Internet have shown, it is possible for millions of people to use a shared standard that is not the property of anyone.

But then, who really would really use or further develop TCP/IP anyways, considering its lack of benefit to a company's business case.

I don't agree that it's trivial to come up with a system of identifying 57 trillion locations using three words. The idea is simple, but implementing it is not.

Code at bottom does Mongolia to 5 decimal places. In one page. All it needs is a word dictionary to go from word number to actual word, and back again. A 2-way hash map.

Now the thing about posting code is that you need both confidence and humility. Confidence enough to offer it, humility to accept "that's wrong" or "this is better" with good grace

I think my 3rd try below is not terrible .. but I might be wrong.

"Strangely enough, the world does not need to pay for an address presently – but now, it can."

Sure it does. Or do you think establishing and maintaining a complete conventional address system via government bureaucracy or consulting contracts would have been cost-free for Mongolia?

' Or do you think establishing and maintaining a complete conventional address system via government bureaucracy or consulting contracts would have been cost-free for Mongolia?'

Are you sincerely suggesting that Mongolia does not presently have a functioning postal system? And that by using this system, the old system will be completely tossed?

The idea is clever enough, and arguably the sort of thing that a small group of people could do easily, particularly when using a license like the GPL to share it. After all, that is pretty much how the system (BIND, Linux/BSD, Apache) that currently runs much of the Internet came into being.

Obviously they do not presently have a functioning postal system

If you'd read the article, you'd know that while Mongolia has a more or less functional postal system (but why is it so hard to believe a country couldn't?), they don't have a functional addressing system. Why that is makes for an interesting question, but what matters is that they need one to increase their postal functionality.

As someone else pointed out, if the Mongolian government wanted to do this "for free" they'd end up having to pay a bunch of bureaucrats and/or contractors. As it is they can just pay these guys instead, and with the benefit of knowing they've already got a workable system in place.

You don't need to pay, just give it as a class assignment.

'they don’t have a functional addressing system'

Except they do, as seen here if one needs to send a letter to the American embassy -

Street Address:
U.S. Embassy
Denver Street #3
11th Micro-District
Ulaanbaatar 14190

Mailing Address:
Section/Employee name
U.S. Embassy in Mongolia
P.O.Box 341
Ulaanbaatar 14192

What they have are many non-fixed locations to deliver mail to. Certainly this is a challenge, but to say that the Soviets were unable to impose an address system on a city is really stretching things.

Further, the idea that people living an essentially nomadic existence are unable to pick up mail is also stretching things. That a GPS/GlONASS based system would speed up delivery times is undeniable, of course - and that a three word unique string would be useful is also fairly plain. On the other hand, the idea that a Mongolian grid would need 9 m2 accuracy is hilarious.

"What they have are many non-fixed locations to deliver mail to. Certainly this is a challenge, but to say that the Soviets were unable to impose an address system on a city is really stretching things."

The problem in many Asian countries is that any address system that have existed in the past cannot keep up new (and often illegal) development. Cities, towns and villages are full of back-alleys, footpaths, and unnamed roads and with no logical building or lot numbering system. The U.S. Embassy in any capital city is typically a prominent building located on a major road in one of the more orderly parts of town so it will rarely be a problem for them to receive mail.

"what3words claims to own all of those 3 word combinations"


Here is one quote from the company, already posted above - 'At the moment, the core what3words algorithms and data are not in the public domain.' http://what3words.com/pricing/

And here is what they mean by 'data' - 'Each what3words language is powered by a wordlist of 25,000 – 40,000 dictionary words.* The wordlists go through multiple automated and human processes before being sorted by an algorithm that takes into account word length, distinctiveness, frequency, and ease of spelling and pronunciation.

Offensive words and homophones (sale & sail) have been removed. Simpler, more common words are allocated to more populated areas and the longest words are used in 3 word addresses in unpopulated areas.' http://what3words.com/about/

If, instead, all we are talking about is generating three word strings to uniquely identify a 9 m2 based grid over the earth, one would not require anything except that generated list - one that could be included in any smartphone with a GPS receiver, so that everyone could exchange those strings, without requiring any pricing considerations at all.

In other words, that company could create a one-off database and distribute it to anyone interested. But in this case, the question mark in that old three part joke about business plans appears in the third position, the place where the word 'profit' normally appears as the punchline.

This strikes me as highly inefficient. I grew up in a town with streets with names (Main Street, Newberry Street, etc.). Later, after I moved away, the town switched to a grid system, where all streets have a number, those north and south called streets and those east and west called avenues. It made finding one's way around very easy. I still use the old names since that's what I remember growing up. When I tell someone I am on Newberry, they are totally confused about my location. Anyway, it seems that a more efficient system for identifying a place would be based on a grid system. You know, like longitude and latitude.

The company is not interested in efficiency, it is interested in its business case.

"Anyway, it seems that a more efficient system for identifying a place would be based on a grid system. You know, like longitude and latitude."

This *is* a grid system based on longitude and latitude -- but instead of having to read/write/remember numbers like 42.23237N, 83.12811W as an address, you use an easy-to-remember 3-word combination.

" In principle, addressing could be done with latitude and longitude but that’s like trying to direct people to web sites with IP addresses–not good for humans."

It's a system for computers (or at least databases) not for people. We are rapidly moving towards a world that is optimized for electronic minds, not biological minds, and this theme (contra Tabarrok) is absolutely terrifying.

Everything old is new again.

I don't understand - doesn't this just replace two numbers with three words?

Also, do the names make help me find one place if I know where another place is? For example, if need to get to 220 Main Street, and I only know where 100 Main Street is, this actually gives me the information I need to know to get to 220 Main Street. Lattitude and Longitude work the same way - if I am at 40 degrees north I know how to get to 41 degrees north.

Does this system have that same kind of logic?

"Lattitude and Longitude work the same way – if I am at 40 degrees north I know how to get to 41 degrees north."

Well, right, but the difference between 40 and 41 degrees is about 140 miles. At the local street level, you're dealing with differences in hundred thousandths of a degree. And, of course, streets are not necessarily oriented along North-South or East-West axes. There's simply no way you could give people 'addresses' consisting of two 7-digit coordinates and think they're going to use those for mental navigation.

I still consider this an undergraduate problem in computer science.

Within the US one latitude or longitude second is about 100 feet. So t get within 20 feet you need .2 second graduation, or about .0006 degrees. Assume Mongolia is similar.

How many degrees does Mongolia span? How many words does it have? From there you get the Mongolian words in your spatial address.

On that we agree. If you aren't worried about reproducing the claimed tweaks for removing homonyms, assigning more memorable combinations to more populated areas, etc, I'd consider the basic task of creating software to translate grid squares into unique 3-word combinations no more than a fun weekend project.

To make it a little more legible, I'd probably make it hierarchical, where the first word would identify a large square, the second word a smaller one, and the third word down to the address level. That way, neighboring properties would have similar addresses. And maybe 4 levels would be better than 3 (you could use a smaller dictionary -- the 4th root of 57 trillion is ~2700). And the pattern would be the same at each level of detail, so you wouldn't need a gigantic lookup table (the same 2700 word lookup table would be used at each level of hierarchy). But, yes, a pretty simple programming project.

> ... I only know where 100 Main Street is ...

This is for places lacking named streets.

> does the system have that same kind of logic?

The system includes a GPS device. Adjacent addresses are forced to be very dissimilar by the algorithm. This is a design goal: minimize mis-deliveries. 83.12181 is easier confused with 83.12151 than flax.wombat.parachute is with mystery.canvas.teacup.

Dissimilar names might not be a good feature for a postal system though. Does not help with route sorting. Indeed, requires computer route sorting.

Using numbers here might even be good. Create nested named boxes on the map, and then sub-divide into 100 addresses per box. If I am at 87 six coyotes dance, route to dance, then coyotes, then six, then 87.



The location name of where they thought this up is: dumb ass idea.

I can see where this is going: vanity addresses in three words: eat.the.rich

Basically Mongolia, trust your smart people to create a Mongolian solution. Don't buy it.

Ireland trusted their smart people to create an Irish solution ...


Interesting that it makes the same non-routing decision as what3words

"However, the random nature of the code has come under fire for being too complicated. Houses in the same area do not have numbers in sequence. Saying the Government “has got it wrong”, Ryan says the codes cannot be easily remembered."

Many mappings are possible, but something meant to say "here I am" to someone else in chat is pretty different from something used to create a delivery list, or to select polling place.

There is a reason traditional addresses are hierarchical.

There's no reason a system like What3Words could not be made hierarchical, where the first word identifies a very large grid square and each succeeding word a smaller grid within the previous square. They do have a rationale for not doing this:

"The what3words system is non-sequential and non-hierarchical to ensure human communication errors are intercepted. The system distributes similar 3 word addresses far apart, often in different countries, to allow manual or automated error detection in real time."

And I get it, but I suspect the drawbacks of assigning words to square quasi-randomly are probably greater than the benefits of error detection.

The non-hierarchical nature is the biggest problem I have with it. And, I think it will increase errors, not reduce them.

The problem is that a small error in any of the words will send you to a totally different part of the world.

Actually, that is key. what3words is optimized for the temporary location of mobile entities. It is not optimized for stationary locations. I can say I am at Throat-Wobbler Mangrove. That's fine. Better to say my house is at USA.CA.X.Y.Z

Right. If this new system is going to be useful (and I'm far from convinced that it will be), it'll be for helping to find mobile locations.

But for real-life navigation, good old fashioned street addresses still have an advantage. E.g. if I'm at 20 Main Street and you're at 5501 Main Street, I know that I need to drive or bicycle or take a bus or Uber or whatever to get to you. Whereas if all I know is that you're at eat.pray.love then I need at a minimum to push a "directions" button to start figuring out what transportation mode I want to use.

I met people at a graduation this past Sunday, and then made sure they could find the restaurant. Each time I pressed the little location circle on my phone, and it sent a lat/lon in a txt message with 7 digits past the decimal point. I didn't need to say anything. They didn't need to do anything other than tap on it. It was interoperable between Apple and Android (finally!)

While latitude and longitude offer helpful coordinates, wouldn't "altitude" be a necessary coordinate for addresses in urban settings?
One-hundred story towers will have multiple locations with identical longitudinal/latitudinal identifiers, no?
How well will this system work for caves and caverns, btw? tunnels and subway systems?

Browsing around my neighborhood I found an address that reads "hangs.common.whites."

I wonder how many other combinations there are that people may not want to speak out loud. (Trying a few of the more obviously problematic prompts fortunately didn't turn up much.)

Yeah, I think my system might just use animal names and happy verbs/adverbs.

Its always fun to watch people claim your innovation is dumb or simple, or so easy. Then, again they didn't think of it or do the work. "You just removed the legs and made them assemble later! That's nothing!" Except it completely revolutionized furniture and allowed mail order furniture to flourish. "You just put standardized metal boxes on a ship! That is trivial!"

Also, people must be under the assumption that Mongolia is a large country with a deep pool of intellectuals. They also assume that because the US embassy in the capital city has an address, well, everyone must have an address!

Now, maybe Mongolia could negotiate a better deal - I didn't look to see how much they are paying.

But if you think about it, the tech company should have a better price than if Mongolia created the system by themselves, because they can sell this system to multiple customers.

Mongolia is smarter to spend its money on this, and focus more on creating mining engineers.

Stupid is as stupid does.

So I took a few minutes and made a python program that converts an xy grid into words and then back again. I am too lazy to clean it up, but proof of concept is very short.


output is transposed, written vertically

87 23
7 3
8 2

but in a simple test case 87,23 is made into word73,word92 and then back into 87,23, written upside down and backwards

better: https://gist.github.com/anonymous/5ab09c4738ca5f931deca60e689ac555

Best: https://gist.github.com/anonymous/581facd452facb1bf7bd11f7ecb03b2b

As noted above, this naming system is not all that good in several real world examples, particularly those involving postal addresses.

As further noted above, this idea is not exactly unique.

'Also, people must be under the assumption that Mongolia is a large country with a deep pool of intellectuals.'

Nope - but I'm sure that a talented high schooler with access to just about any internet connected PC could create a database populated with three word strings that detail 9 m2 plots, and that this database could be easily distributed and used for addressing. This is a fairly trivial thing to do - and currently, all the tools necessary to do it can be found on the Internet, for free.

At worst, Mongolia could see if any Cyrillic using Russian programmers are interested in earning a quick couple of hundred dollars/euros/etc. to do the work.

'They also assume that because the US embassy in the capital city has an address, well, everyone must have an address!'

Not everyone has an address in a society notable for it nomadism, which is not the same thing as saying Mongolia has no addresses at all. Places with addresses would undoubtedly include pretty much any fixed town/mining camp/medical facility/fuel depot/etc.

'Now, maybe Mongolia could negotiate a better deal – I didn’t look to see how much they are paying.'

As pointed out above, Mongolia need not look for a better deal at all - the government need only find a single class where students are capable of using Internet connected PCs to implement this system as a class assignment.

'should have a better price'

You are unfamiliar with the GPL, right? Even though without it, the Internet would not exists as it does today. In large part because no one needed to pay anyone the right to implement/use TCP/IP, or BIND, or HTTP, etc.

'Mongolia is smarter to spend its money on this, and focus more on creating mining engineers.'

Who, hard as this might be to imagine, will also have to know something about software to be professionally useful at all. This is truly the sort of thing that an intro to programming class at the freshman level would be expected to handle - again, this is nothing more than attaching 3 word strings to plotted locations.

The system relied too much on GPS and it is easily manipulated by private or state actors.



In a court case a accused can swear that he, according to his GPS smart watch, he was not at location x.y.z at time t and he was not familiar with the said location. Can a eye witness account about the location be trusted? Since the spoofing error rate is nonlinear and is depended on the location and altitude, the x.y.z and t wrt to the accused might not even be the same as the x.y.z and t wrt to the witness. There could even be more than one spoofing devices around that region thus the two set of data might not be able to be correlated.

You've got one of the best online sites

Comments for this post are closed