I didn't see any discussion in the article of having different languages but the what three word website talks about it.
One thing that I don't quite understand is two locations very near each other have completely different sets of words. I would have thought there would be some sort of hierarchy of words (sort of like geocoding a lat long where it defines a box and the precision of the geocoding gives you larger or smaller boxes).
I wonder too if there are going to be collisions of words and locations that aren't going to make people too happy. Though they must have thought through that too.
After trying some locations, I think they should have purged their dictionary of obviously negative words. Ending up with `glorified.bodily.passage` is just bad luck and can't be avoided, but `snored`, `brainwashed` or `grieved` are just bad in any context.
Also, some ground offset in meters (grazed.across.like.7) or the ability to go beyond 3x3 meters like a fractal would allow you to specify individual drawers or even pencils inside them.
If Wikipedia is to be believed, that's intentional as an error-detecting mechanism:
"The result is that if you enter athree-word combination slightly incorrectly and the result is still a valid w3w reference, the location will be so far away from the user’s intended area that it will be immediately obvious to both a user and an intelligent error-checking system."
I'm skeptical of just how well that'll work, personally. Witness the example of the truck driver who drove to Gibraltar, England instead of the one next to Spain.
"Thus far, the system has been built in 10 languages: English, Spanish, French, German, Italian, Swahili, Portuguese, Swedish, Turkish, and, starting next month [article is from Nov 2015], Arabic"
Except the alternative is probably even worse, so it might be a net reduction in confusion - here in Costa Rica my address is a lengthy sentence of directions from a locally well known school nearby!
Amazon Global will deliver most things right to my door now and it's surprisingly inexpensive shipping + customs taxes (in some categories). Hard part is finding stuff that ships directly, they have a checkbox to filter for international shipping but it's not completely accurate. Shipping time is 2 - 3 weeks minimum.
For everything else and stuff Amazon doesn't ship I use Aerocasillas or a PO box at the post office. Aerocasillas are pricey but they do their job and fast, something arriving at their address in Miami on Monday is usually available to pickup by Friday. The post office recently didn't tell us about a package my sister sent until after they'd sent it back lol.
I was actually reasonably impressed with the system; they don't put nearby words near each other, so if you have a typo it becomes pretty obvious to anything but the most braindead delivery system.
So if what you're writing on the envelope is
Bob Smith
reporter.smoked.received
France
it becomes trivial to correct that to reporter.smoker.received, unlike when you write "124 Main Street" instead of "123 Main Street".
To what3words' credit, when I typed reporter.smoker.received into the map tool, it did suggest the correct location in its list of similar locations. https://imgur.com/G2ChrxA.png
Right. What happens when you call 911 and they send the ambulance to "gravy.sun.request" instead of "gravy.sun.requests" or whatever?
The odds of such a similar name being located in the same city are low, so they'd probably notice the discrepancy, but there's still a lot of room for error in communicating the words, since they're not normalizing singular/plural or tenses.
I can think of two phases. First, the problem is with people who don't have an address. They currently have to say "send an ambulance to the blue building just above the X gas station near the police station in west Y". Having some sort of address - even with ambiguity! - is better than now. And we don't know if there is local ambiguity.
(Also, dialects make a difference. My dialect has "pen/pin" as a homonym. Rhoticity adds another layer of complications.)
Then again, we already have homonym problems. For example, "Greene St." vs "Green St." Newark, Jersey City, NYC, and Brooklyn each have one or the other, within a span of a few miles. For the ambulance problem, E911 system in the US has a rough idea of where you are calling from, and can help disambiguate.
Which means we know it's possible to build a system on top of similar names. But the first step is to get some names.
Yes, three-word phrases have more cues for memory. They mean more than a random string of numbers. It also looks similar to a method for improving memory, by finding weird mnemonics.
Weird is memorable, and more meaning acts like a checksum to reduce errors.
Especially English meaningless combinations for 3 words. If you don't know English it might as well be a set of numbers, except you'd also have to memorize the 26 letter alphabet.
The correct solution to a transcription error is a checksum algorithm. See for example the implementation I made for reverse-searching incorrect IBAN from a liberal list of probable transcription errors @ https://github.com/globalcitizen/php-iban/ .. it's very fast, cheap and accurate.
If you read eg cyrillic you're going to find it very much easier to read and remember cyrillic words. Thus, presenting cyrillic to those people benefits them. That's why translating from English to local languages is a good thing.
No, they don't seem to have been literally translated. My Swedish address is not the translation of my English address.
Which is kind of interesting in its own right, since you can then refer to the same physical location in different ways depending on which language you use.
This calls for some kind of geographic cross-lingual poetry.
Carefully vetted? I don't have the time right now but there are countless weird, insulting, offensive phrases possible with this proprietary abomination.
Unless they went back and manually deleted words after their second round of hype last year, they are still in. Their wikipedia page surely has been largely written by them themselves.
The article lists Mongolian US embassy address as "constants.stuffy.activism", except the map they link to gives me the address "indoor.sheepish.blues". How can it be an address if it changes with time?
It doesn't. It's a grid system with 3 meter squares, each one having its own trio of words. Move the point 3 meters in any direction and you'll get a different one. This allows much more precise and close together addresses, down to the entrances of doors or gates.
You still pay for writes(i.e. domain registration). You can argue that DNS and domain registration are different things but in reallity they are inter-dependent so there is still a cost and the system is "owned" by several entities(i.e. ICANN)
What3Words is free for writes (presumably it's generated from an algorithm and there aren't really "writes") and you pay for reads.
Kind of a big difference. With DNS you only have to pay if you want a domain. I can freely do a DNS query to resolve "flamingmongoosefarts.com", but with What3Words I (might not) be able to freely determine where flaming.mongoose.farts is.
DNS query is free because someone pays for the domain registration and your DNS queries(DNS queries/servers cost too). I can see how What3Words could charge the address owner a fee for its address to be licensed publicly just like the domain registration publishes your NS records as long as the maintenance fee is paid.
Tbh I must admit that they are slightly different scenarios because the domain name registration is slightly distributed but my point is that public DNS/Domain registration is neither free nor open. They are controlled by various entities/corporations so they are not the best examples.
"You shall not use the API for any application that replicates or attempts to replace the essential user experience of the Service or use the API to create a Product which functions substantially the same as the API and offer it for use by third parties. You must not pre-fetch, cache, index, copy, re-utilise, extract or store any what3words Data"
I really don't think it's a smart move to have a national entity totally depend on a private company that has no plans to make its database freely accessible for its operations; the worst being that said private company is foreign, meaning that Mongolia has 0 leverage on how the database will evolve, it is going to be completely at the mercy of What3Words' changes of policy and access rights.
Unless What3Words' database is going to be fully accessible with no restrictions for everyone, this really looks like a bad idea.
I guess they might have to break some treaties or something to do that, but a country isn't going to be beholden to a small foreign company for much of anything.
Mongolia has expressed limited sentiments towards joining TPP but they try to remain neutral between the various regional and global factions, and being a member of TPP could irritate their Russian and Chinese partners.
TTIP is similar, but you might as well just say TTIP if you mean TTIP and not TTP. Also, even if the UK is partner, Mongolia certainly can't be, at least to this particular partnership.
I don't disagree with what you are getting at, but there is a difference between it being in their best interests to maintain their relationships and the company having enormous leverage over them. I think even if What3Words were a US company, the US gov. wouldn't do a whole lot about it if Mongolia just decided to stop paying them and use the locations anyway.
I completely agree. Seems like the underlying technology wouldn't be that difficult to reimplement independently, or am I mistaken? Surely a few CS interns in Mongolia would be up for slapping together a vital national infrastructure project on the government's dime, eh?
Unless your yard is less than 3 square meters, you'll have more than one phrase for your address. Enumerate all the possible phrases, and choose the coolest one...
No it replaces city, state, country too. (Yes, you might need to include some of those to be backwards compatible with the current mailing system, but that's not how this system would work.)
I would add to the other comment that there are plenty of ways this doesn't improve your life in San Jose. But the Vox article points out those 70K people who live in a slum of Rio do have the benefit of a working system.
I feel like it could definitely be made more user-friendly, though. Like instead of a random assortment of 3-word combinations, is there no way to assign them in an ordered fashion, so that "hill.stevedore.borg" is west of "hall.stevedore.bar". Or use a larger grid with an inset numeric system, like instead of blocking in 3 square meter chunks, use 30 square meter chunks with a 2-digit number next to it, like "hambone.impish.doorframe-3,3".
I kinda feel like any arbitrarily-ordered locational identifier will always feel clunky compared to an ordered grid system.
Address is pretty separate from directions in both the old and new system. The fact that I'm on Maple st. doesn't tell me anything about how close I am to Pine st., except in a small number of cities that have a useful naming system. So I say "Turn left on Maple, then right on Pine, then...". Likewise, here you say "turn right at maple.big.grumpy, then left at green.thick.rat".
This system is an improvement for locations with no road names (so it's not any harder to give directions) or with ubiquitous GPS (where you can directly convert). The number of places with road names but not ubiquitous GPS is on the fast decline.
Maybe I am just dumb but I cannot really understand the "Address is pretty separate from directions in both the old and new system".
You usually know where you live (i.e. how to get there) so the only reason you need to have some kind of coordinate system (be it "conventional" or the one we are discussing) is to tell someone other, including the postman, how to get to you.
I.e. - I really do not understand why "Hi Ibrahim, I live at Alpha-Unicorn-Honey, just drop here when you are in the area" would beat "The train stops at Nadira, from there you walk 2 miles in the direction of the sunset, when you reach a rock looking like a kneeling woman you turn in the direction of the face and walk anoter mile - and there is my hut".
So far the only place where I believe this system would help is in large Japanese cities (Tokyo) etc. because they don't have "named" streets, most of the time, and at the same time most people can be expected to have access to technology.
(Maybe it's the same in China, but never been there so I don't know how it is)
I sincerely doubt that people living on the outskirts of the Gobi desert will routinely exchange their three-words coordinates over whatsapp and then get Google Maps or whatever to guide them to the desired destination.
> I really do not understand why "Hi Ibrahim, I live at Alpha-Unicorn-Honey, just drop here when you are in the area" would beat "The train stops at Nadira, from there you walk 2 miles in the direction of the sunset, when you reach a rock looking like a kneeling woman you turn in the direction of the face and walk anoter mile - and there is my hut".
You misunderstand. It's not trying to be better at directions, it's trying to be better at unambiguously specifying a location. If you're mailing something to someone where there are no roads, then this gives you a way to address them with more precision than you could do before. You just need a postman with a $40 smart phone with GPS (or a look-up-table).
You should really think about this as "easy to remember and communicate GPS coordinates".
The idea is quite interesting, but the implementation only makes sense for sparsely-populated regions. An apartment in the city with two side-by-side doors to different units will easily fit within a three-meter square, meaning that you'll still require "Apt. #1, Apt. #2" designations to disambiguate. Similarly, the fact that this system accounts only for geographical surface coordinates means that any building with distinct addresses stacked vertically, like an apartment building, won't be able to be served by this system alone.
Apartment units addresses will probably be based on the 3x3 square the most closely matches its outline, so door location will be irrelevant. Vertical stacks can be simply dealt with by assuming a standard 3 or 4 meter story, so unless you live in an industrial loft or something it won't be a problem.
Door location is how you can tell if a house is in the Netherlands or in Belgium, for Baarle-Nassau/Baarle-Hertog. Move the door and you move countries.
Outline is not time invariant. If I turn a duplex into a single family, or build out, then the shape changes. If I live in Coober Pedy and build out underground, then do I use the surface outline or the underground outline?
If I relocate the house a few meters away, the address definitely changes.
Really, this is best for the billions of people who have no address. Most likely it will boot-strap, so in 100 years the name "X.Y.Z" will refer to the ghost of the place that once was, but everybody local knows what it now means, as will the databases. With a layer of apt. #1, #2, etc. on top of that.
It doesn't depend on sparsely population regions, which is why it is also being used for Ulaanbataar.
It depends on being better than the current system (which is description based), and better than the alternatives (like putting up street signs everywhere).
One problem I see is that one particular address may end up with several 3-word addresses. Since physical distance doesn't affect the similarity of the combinations, how is one to determine the 'correct' combination for a particular location? This seems to compound the inherent problem of all 3-word combinations sharing the same namespace - I have no idea if two addresses are actually the same building, or on opposite sides of the planet.
For most things the adjacent addresses would be equivalent. It doesn't matter which part of the house a package is delivered to. In many cases someone picking an address won't really be able to get within 3 meters anyway.
I guess people looking at several addresses won't be looking at the words, they will have to have a resolver and look at them that way.
This remains a solved problem. The location phrase specifies the coarse location. Additional addressing information is used to introduce granularity. Look at how addresses are conventionally written already, and in a good chunk of the world, the addresses are written top to bottom, most specific to least specific. The largest remaining problem is that the location mapping depends on a proprietary database, but this is a common problem (in Canada the database of postal codes is also proprietary).
that's mostly true, except US/Canada we put the zip/postal code at the end which is more specific than the city/state/province. But your gist remains true.
Presumably this is handled by the software that is being licensed. Because if all the software does is do a conversion from words to gps coords, they're being taken for a ride.
40,000 words gives 15.29 bits per word, for a 45.86 bit locator. log2(area of earth / 9m^2) is 45.7.
Packing bits into phrases is also useful for passwords (https://rmmh.github.io/abbrase/), but in my experience the increased difficulty of correctly remembering phrases from a large wordlist outweighs the additional entropy.
4 words chosen from a 2800-word list would convey the same amount of information, and be less prone to confusion.
It looks like they're not attempting to make the triplets have consistent meanings between languages, which makes sense -- 40,000 word correspondences between 10 languages would be exceedingly difficult! logs.broker.remote is sweight.saum.zulässige is lodos.pasado.ninguno, none of which are the same meanings.
Another improvement would be a hierarchical system-- you can specify a more coarse location using more common words -- picking from a wordlist of 500 for a 4km^2 specifier, 2000 for for a 16 acre plot, 10000 for a 500m^2 area.
> Packing bits into phrases is also useful for passwords (https://rmmh.github.io/abbrase/), but in my experience the increased difficulty of correctly remembering phrases from a large wordlist outweighs the additional entropy.
I saw a paper with a nice method for generating passphrases. Unfortunately, I can't find it. (Can anyone help me?)
First, you generate random data with enough entropy for a passphrase. Then, instead of encoding it in a sequence of words that is as short as possible, you encode it in many (say, a million) sequences that are a bit longer. Each sequence thus encode the same data. Now, you feed all these sequences through an algorithm that gives each sequence a score depending on how close it is to being a natural-language phrase. Finally, you present the user with the sequences, ordered by their score, and let the user choose any one of the to use as passphrase.
I don't know that paper, but that's similar to what's going in Abbrase. It shows the best-scored mnemonic using bigram and frequency heuristics, but you could show more alernatives to let users choose their own mnemonic.
You can see multiple alternative mnemonics using `abbrase.py -m` -- it hasn't been implemented in the web version yet. It does give better results, from "wave several ominous silence" (default based on heuristic score) to "waves sever ominous silk" or "wavy seven omitted silver" (human-chosen mnemonics).
> I saw a paper with a nice method for generating passphrases. Unfortunately, I can't find it. (Can anyone help me?)
DiceWare[1] doesn't do the filtering you're talking about, but it does provide a method of generating passphrases by converting entropy into words. PGP Word Lists[2] use a packing technique as well, with even/odd words for error detection.
I haven't seen the paper you're talking about, but I'm highly suspicious of using an algorithm to score phrases by how close to natural language they are without a proof of security.
If you score natural-ness using, for example, comparing overlapping 2-word patterns to a database, then that database becomes a Markov chain database we can use to generate passphrase guesses, greatly reducing the entropy. Generating and discarding the patterns database from natural language material for each passphrase generation helps only inasmuch as two source databases from the same language are likely to differ. Maybe that's not the method used to test natural-ness, but I suspect that the better the natural-ness scoring function is, the better the corresponding passphrase generator will be. And remember, the user is going to decrease entropy even further by choosing their passphrase.
You can mitigate this by decreasing the number of passwords generated for the user to choose from, but this means less natural phrases. You can also mitigate it by increasing the passphrase space (that is, more words in your passphrase) but that decreases memorability.
Personally, I use 5-word DiceWare passphrases for most things, and 6-word DiceWare passphrases for the most important stuff. The proof of security for DiceWare is simple enough that you can do it yourself and be confident in it. The intent is not long-term memorability--in the age of leaked passwords stored or transferred in plaintext, password reuse is foolish, and I've yet to come across a scheme that allows me to memorize 30+ passwords. Instead the idea is short-term memorability so I can copy the passwords manually from my password manager. In practice the memorability is completely adequate for this purpose.
And even at 5 DiceWare-generated words, that's 2.8x10^19 possibilities, or a little under a year for an attacker with 1 trillion guesses per second capabilities.
> I'm highly suspicious of using an algorithm to score phrases by how close to natural language they are without a proof of security.
To clarify, all the phrases to be scored represent the same random data. They are just more or less difficult to remember. The natural-language algorithm is just there to help you find one that is easy to remember, and is not security critical. The algorithm is assumed to be known to be known to the attacker in any case.
So do you live at peak.pique.peek, or peak.peek.pique? Or is it... What floor? Oh, you're on the top floor? So you leave on the peak of pique.peak.peek?
My guess there are 100s miles between anything that is similar (typos, same words but different order) i.e., you'll be able to figure out from the context which address is correct.
This is absolutely hilarious, at least to my inner six-year-old. They seem to have filtered out words with obvious negative connotations like "ugly", words of under four letters, and maybe longer words. But through the magic of language, even very innocent words are subject to reinterpretation. And a lot of What3Words's words are actually not that innocent.
for the Marvel fans, longing.rusted.furnace is a mountain road north of Verona, Italy. The system doesn't use numbers, so thats as far as you can go if you want to write some sort of Winter Soldier fanfiction based on this.
in French, they probably put the most unused words they could find in the dictionnary. As an example, you have 'maussaderie', 'taniser', 'houblonner', 'époussette' or 'numismate'. I'm sure a good proportion of native speakers would have to use a dictionnary to find the meaning.
Even a small house has dozens of addresses. I checked and there is a different address in my kitchen than my bathroom, based on what3words website. If you find a certain combination offending, you can choose a different one, but on the other hand, on a single property there could be thousands of addresses, many of which funny or embarrassing.
Or you just use geocoordinates like everyone else. This project is aimed to make addresses more memorable for humans, machines do not have such a problem.
Machines do not. However, humans would find it difficult to memorize the GPS coordinates for their front porch - which is why trying an approach like this with pure GPS might not work.
It's interesting that this is so fun. (My home address resolves to something satisfyingly naughty.) An app that just generated random three word phrases wouldn't be fun, and neither would one that mapped every point on earth to a dirty phrase.
There was a really cool project using syllables such as consonant+vowel, then using two "words" with 3 syllables each such as:
"tapiku norami"
But the cool thing is that they were basically real gps coordinates just encoded as letters, so returning them to the original was trivial. I think a couple of letters were avoided to avoid confussion.
Unfortunately I cannot remember the name and they did a terrible SEO job...
"alien.crash.landing"[0] is a remote farm field in Montana. Just the kind of place you'd expect.
"lower.life.expectancy"[1] is some patch of land in Colorado. I can't help but notice the two man-made pits containing unknown substances a few hundred meters north.
Eircodes were created that way so that people wouldn't have to change their address at all. Which could be a politically difficult thing. Politicians choose the thing that made their life easier and stopped complaints.
So people in Mongolia will use English words for addresses? The whole "easy to remember" is completely lost when it's a foreign language.
Imagine if America switched to three Chinese characters for an "easy to remember" address system.
In China numbers are often used inside of roman character email addresses, or domains, because roman characters are only convenient for westerners, much less English words.
See 1688.com (alibaba.com), or use of QQ numbers for email addresses.
> Imagine if America switched to three Chinese characters for an "easy to remember" address system.
This sounds pretty unlikely even for China, since it's almost impossible to unambiguously name any particular character. When speaking, you would generally identify a specific character using a four-syllable phrase like 安静的静 (literally "the jìng of ānjìng [quiet]"). Why can't you just say jìng? Well, here are some other characters with the same pronunciation:
竟 境 镜 敬 竞 净 径 靖 劲 痉 胫 婧 獍 靓
There are a few characters with unique pronunciations, like 给 or 谁, but those are very much not the norm.
The chinese version of what3words would be overwhelmingly likely to use words, not characters.
Singapore is a tiny island but it's system is Super easy. It uses post codes assigned to physical buildings. From a post code you can get the name of the building, street, and access.
190 comments
[ 3.4 ms ] story [ 209 ms ] threadOne thing that I don't quite understand is two locations very near each other have completely different sets of words. I would have thought there would be some sort of hierarchy of words (sort of like geocoding a lat long where it defines a box and the precision of the geocoding gives you larger or smaller boxes).
I wonder too if there are going to be collisions of words and locations that aren't going to make people too happy. Though they must have thought through that too.
Either way it is an interesting idea.
If you don't like the 3 words it gives you, just offset the location by a foot or two until you find a set of 3 words you like.
> If you don't like the 3 words it gives you
After trying some locations, I think they should have purged their dictionary of obviously negative words. Ending up with `glorified.bodily.passage` is just bad luck and can't be avoided, but `snored`, `brainwashed` or `grieved` are just bad in any context.
Also, some ground offset in meters (grazed.across.like.7) or the ability to go beyond 3x3 meters like a fractal would allow you to specify individual drawers or even pencils inside them.
MGRS is (approximately) a dense encoding of UTM coordinates, at similar precision it's a lot less memorable than 3 words:
4QFJ 12345 67890
(from https://en.wikipedia.org/wiki/Military_grid_reference_system )
But it does have the property that coordinates are easy to compare to each other. Less precision can be used just by dropping digits:
4QFJ 1234 6789 - the 10 meter zone that the 1 meter zone above falls inside
4QFJ 123 678 - the 100 meter zone
"The result is that if you enter athree-word combination slightly incorrectly and the result is still a valid w3w reference, the location will be so far away from the user’s intended area that it will be immediately obvious to both a user and an intelligent error-checking system."
I'm skeptical of just how well that'll work, personally. Witness the example of the truck driver who drove to Gibraltar, England instead of the one next to Spain.
http://qz.com/561727/an-address-for-every-place-in-the-world...
Also:
"Each square has its identity in its own language that is not a translation of another."
For everything else and stuff Amazon doesn't ship I use Aerocasillas or a PO box at the post office. Aerocasillas are pricey but they do their job and fast, something arriving at their address in Miami on Monday is usually available to pickup by Friday. The post office recently didn't tell us about a package my sister sent until after they'd sent it back lol.
> the Stade de France is at reporter.smoked.received
The phrase they mention goes to some place in Trenton, Missouri[1]. It seems they meant reporter.smoker.received[2].
[1] https://map.what3words.com/reporter.smoked.received
[2] https://map.what3words.com/reporter.smoker.received
So if what you're writing on the envelope is
it becomes trivial to correct that to reporter.smoker.received, unlike when you write "124 Main Street" instead of "123 Main Street".Are we sure? They could do something like a Levenshtein distance correction, but they also might not have.
Are we sure? They could do something like a Levenshtein distance correction, but they also might not have.
The odds of such a similar name being located in the same city are low, so they'd probably notice the discrepancy, but there's still a lot of room for error in communicating the words, since they're not normalizing singular/plural or tenses.
(Also, dialects make a difference. My dialect has "pen/pin" as a homonym. Rhoticity adds another layer of complications.)
Then again, we already have homonym problems. For example, "Greene St." vs "Green St." Newark, Jersey City, NYC, and Brooklyn each have one or the other, within a span of a few miles. For the ambulance problem, E911 system in the US has a rough idea of where you are calling from, and can help disambiguate.
Which means we know it's possible to build a system on top of similar names. But the first step is to get some names.
Weird is memorable, and more meaning acts like a checksum to reduce errors.
(that is, they are arguing that most people will have an easier time remembering words from their native language)
If you read eg cyrillic you're going to find it very much easier to read and remember cyrillic words. Thus, presenting cyrillic to those people benefits them. That's why translating from English to local languages is a good thing.
Which is kind of interesting in its own right, since you can then refer to the same physical location in different ways depending on which language you use.
This calls for some kind of geographic cross-lingual poetry.
Kind of a big difference. With DNS you only have to pay if you want a domain. I can freely do a DNS query to resolve "flamingmongoosefarts.com", but with What3Words I (might not) be able to freely determine where flaming.mongoose.farts is.
"You shall not use the API for any application that replicates or attempts to replace the essential user experience of the Service or use the API to create a Product which functions substantially the same as the API and offer it for use by third parties. You must not pre-fetch, cache, index, copy, re-utilise, extract or store any what3words Data"
Unless What3Words' database is going to be fully accessible with no restrictions for everyone, this really looks like a bad idea.
I guess they might have to break some treaties or something to do that, but a country isn't going to be beholden to a small foreign company for much of anything.
Mongolia has expressed limited sentiments towards joining TPP but they try to remain neutral between the various regional and global factions, and being a member of TPP could irritate their Russian and Chinese partners.
Mongolia depends on foreign aid. They'll think thrice before angering Uncle Sam, justified or no.
I don't disagree with what you are getting at, but there is a difference between it being in their best interests to maintain their relationships and the company having enormous leverage over them. I think even if What3Words were a US company, the US gov. wouldn't do a whole lot about it if Mongolia just decided to stop paying them and use the locations anyway.
https://patentscope.wipo.int/search/en/detail.jsf?docId=WO14...
a) suppose I am in Apt #312 in Bellevue Mansion, Whatever Street 23, San Jose, California.
Apt. #311 is on a wildly different set, so how can I find out we are neighbours? Or how can I explain to someone where Whatever Street #21 is?
b) what if I have the same 3-words square, but you are on floor #7 and I am on floor #11?
hellow.world.hola Floor 7 Apt 781 City, Zip, ST, Country
I kinda feel like any arbitrarily-ordered locational identifier will always feel clunky compared to an ordered grid system.
This system is an improvement for locations with no road names (so it's not any harder to give directions) or with ubiquitous GPS (where you can directly convert). The number of places with road names but not ubiquitous GPS is on the fast decline.
I.e. - I really do not understand why "Hi Ibrahim, I live at Alpha-Unicorn-Honey, just drop here when you are in the area" would beat "The train stops at Nadira, from there you walk 2 miles in the direction of the sunset, when you reach a rock looking like a kneeling woman you turn in the direction of the face and walk anoter mile - and there is my hut".
So far the only place where I believe this system would help is in large Japanese cities (Tokyo) etc. because they don't have "named" streets, most of the time, and at the same time most people can be expected to have access to technology.
(Maybe it's the same in China, but never been there so I don't know how it is)
I sincerely doubt that people living on the outskirts of the Gobi desert will routinely exchange their three-words coordinates over whatsapp and then get Google Maps or whatever to guide them to the desired destination.
You misunderstand. It's not trying to be better at directions, it's trying to be better at unambiguously specifying a location. If you're mailing something to someone where there are no roads, then this gives you a way to address them with more precision than you could do before. You just need a postman with a $40 smart phone with GPS (or a look-up-table).
You should really think about this as "easy to remember and communicate GPS coordinates".
Outline is not time invariant. If I turn a duplex into a single family, or build out, then the shape changes. If I live in Coober Pedy and build out underground, then do I use the surface outline or the underground outline?
If I relocate the house a few meters away, the address definitely changes.
Really, this is best for the billions of people who have no address. Most likely it will boot-strap, so in 100 years the name "X.Y.Z" will refer to the ghost of the place that once was, but everybody local knows what it now means, as will the databases. With a layer of apt. #1, #2, etc. on top of that.
It depends on being better than the current system (which is description based), and better than the alternatives (like putting up street signs everywhere).
I guess people looking at several addresses won't be looking at the words, they will have to have a resolver and look at them that way.
Sure, this system doesn't make that much sense in a place like London or SF.
The history of post addresses is quite interesting in it's own right :)
Packing bits into phrases is also useful for passwords (https://rmmh.github.io/abbrase/), but in my experience the increased difficulty of correctly remembering phrases from a large wordlist outweighs the additional entropy.
4 words chosen from a 2800-word list would convey the same amount of information, and be less prone to confusion.
It looks like they're not attempting to make the triplets have consistent meanings between languages, which makes sense -- 40,000 word correspondences between 10 languages would be exceedingly difficult! logs.broker.remote is sweight.saum.zulässige is lodos.pasado.ninguno, none of which are the same meanings.
Another improvement would be a hierarchical system-- you can specify a more coarse location using more common words -- picking from a wordlist of 500 for a 4km^2 specifier, 2000 for for a 16 acre plot, 10000 for a 500m^2 area.
I saw a paper with a nice method for generating passphrases. Unfortunately, I can't find it. (Can anyone help me?)
First, you generate random data with enough entropy for a passphrase. Then, instead of encoding it in a sequence of words that is as short as possible, you encode it in many (say, a million) sequences that are a bit longer. Each sequence thus encode the same data. Now, you feed all these sequences through an algorithm that gives each sequence a score depending on how close it is to being a natural-language phrase. Finally, you present the user with the sequences, ordered by their score, and let the user choose any one of the to use as passphrase.
You can see multiple alternative mnemonics using `abbrase.py -m` -- it hasn't been implemented in the web version yet. It does give better results, from "wave several ominous silence" (default based on heuristic score) to "waves sever ominous silk" or "wavy seven omitted silver" (human-chosen mnemonics).
DiceWare[1] doesn't do the filtering you're talking about, but it does provide a method of generating passphrases by converting entropy into words. PGP Word Lists[2] use a packing technique as well, with even/odd words for error detection.
I haven't seen the paper you're talking about, but I'm highly suspicious of using an algorithm to score phrases by how close to natural language they are without a proof of security.
If you score natural-ness using, for example, comparing overlapping 2-word patterns to a database, then that database becomes a Markov chain database we can use to generate passphrase guesses, greatly reducing the entropy. Generating and discarding the patterns database from natural language material for each passphrase generation helps only inasmuch as two source databases from the same language are likely to differ. Maybe that's not the method used to test natural-ness, but I suspect that the better the natural-ness scoring function is, the better the corresponding passphrase generator will be. And remember, the user is going to decrease entropy even further by choosing their passphrase.
You can mitigate this by decreasing the number of passwords generated for the user to choose from, but this means less natural phrases. You can also mitigate it by increasing the passphrase space (that is, more words in your passphrase) but that decreases memorability.
Personally, I use 5-word DiceWare passphrases for most things, and 6-word DiceWare passphrases for the most important stuff. The proof of security for DiceWare is simple enough that you can do it yourself and be confident in it. The intent is not long-term memorability--in the age of leaked passwords stored or transferred in plaintext, password reuse is foolish, and I've yet to come across a scheme that allows me to memorize 30+ passwords. Instead the idea is short-term memorability so I can copy the passwords manually from my password manager. In practice the memorability is completely adequate for this purpose.
And even at 5 DiceWare-generated words, that's 2.8x10^19 possibilities, or a little under a year for an attacker with 1 trillion guesses per second capabilities.
[1] http://world.std.com/~reinhold/diceware.html
[2] https://en.wikipedia.org/wiki/PGP_word_list
To clarify, all the phrases to be scored represent the same random data. They are just more or less difficult to remember. The natural-language algorithm is just there to help you find one that is easy to remember, and is not security critical. The algorithm is assumed to be known to be known to the attacker in any case.
Ocean squares (and Antarctica) don't appear to be localized -- log2(25000^3) = 43.82 > log2((land area of earth - area of antarctica) / 9m^2) = 43.8.
So do you live at peak.pique.peek, or peak.peek.pique? Or is it... What floor? Oh, you're on the top floor? So you leave on the peak of pique.peak.peek?
glorified.bodily.passage is in some farmland in the Netherlands, near Wijk en Aalburg: https://map.what3words.com/glorified.bodily.passage
beats.member.daily is in a gated community on Long Island: https://map.what3words.com/beats.member.daily
tugging.bunny.organ is a farmhouse or suburban house in Hungary: https://map.what3words.com/tugging.bunny.organ
pigs.pigs.pigs is a vacant lot in Loveland, Colorado: https://map.what3words.com/pigs.pigs.pigs
inflamed.flesh.massaged is some farmland near Stoke-on-Trent in England: https://map.what3words.com/inflamed.flesh.massaged
It's a lot worse in Spanish for some reason.
https://map.what3words.com/needy.member.smoked
https://map.what3words.com/under.water.loans is south of London
https://map.what3words.com/shame.without.regret is in Norway
https://map.what3words.com/regret.without.shame is in Alaska
https://map.what3words.com/winter.winter.winter is near Chicago
https://map.what3words.com/summer.sunshine.happy is in Beijing
https://map.what3words.com/forgotten.previous.husband is in the middle of the south Atlantic ocean
https://map.what3words.com/imaginary.future.partner is somewhere in Russia
etc.
You could literally direct the packet to your front porch, balcony or whatever else place is best for you to accept it.
surely.trim.organ stuff.small.organ tickle.long.organ milk.long.organ milk.needy.organ
glorified.body.openings
stiff.little.finger stiff.large.organ
take.fine.compost
wrong.wrong.wrong
https://map.what3words.com/maybe.another.day
"tapiku norami"
But the cool thing is that they were basically real gps coordinates just encoded as letters, so returning them to the original was trivial. I think a couple of letters were avoided to avoid confussion.
Unfortunately I cannot remember the name and they did a terrible SEO job...
[1]: https://en.m.wikibooks.org/wiki/Lojban/Numbers
> Users can look up these three-word locations, which have been translated into Mongolian, for free on the What3Words website or app.
"lower.life.expectancy"[1] is some patch of land in Colorado. I can't help but notice the two man-made pits containing unknown substances a few hundred meters north.
[0] http://w3w.co/alien.crash.landing
[1] http://w3w.co/lower.life.expectancy
doesn't sound very Chinese government to me. This has a hefty smell of bollocks to me.
Turns out I was confusing that with Tibet. Embarrassing to say the least...
https://stiobhart.net/2016-01-15-stupidest-idea-ever
Imagine if America switched to three Chinese characters for an "easy to remember" address system.
In China numbers are often used inside of roman character email addresses, or domains, because roman characters are only convenient for westerners, much less English words.
See 1688.com (alibaba.com), or use of QQ numbers for email addresses.
I agree with your original point that it is strange that the Mongol Post would want to use English words for addressing.
Edit: they support other languages; I assume they would add Mongolian to the list.
This sounds pretty unlikely even for China, since it's almost impossible to unambiguously name any particular character. When speaking, you would generally identify a specific character using a four-syllable phrase like 安静的静 (literally "the jìng of ānjìng [quiet]"). Why can't you just say jìng? Well, here are some other characters with the same pronunciation:
竟 境 镜 敬 竞 净 径 靖 劲 痉 胫 婧 獍 靓
There are a few characters with unique pronunciations, like 给 or 谁, but those are very much not the norm.
The chinese version of what3words would be overwhelmingly likely to use words, not characters.