What happens when What3Words gets lost in translation? What3Words, the website and app that translates physical coordinates into short memorable combinations of words, has been praised and criticized over the years. Now a computer scientist at the University of Exeter in the UK has formally described in a paper how confusion can arise from the geocoding algorithm used by …
In the UK, the Ordnance Survey's OS Locate app tells you your location, in the National Grid system (you can set it to 6, 8, or 10 digit precision).
Doesn't need a data connection.
Where is the I'd-rather-not-bet-my-life-on-a-foreign-corporation's-proprietary-algorithm-in-an-emergency icon, anyway?
I thought I heard Morgan Freeman say the universe was 15.8 billion years old and was incredulous. Had to rewind. He said 13.8.
Numbers don’t get transmitted flawlessly either.
Yes, people get confused. This is not newsworthy. If I ring 999 and say “help, I’m stuck on snowdonia” but my w3w location is London. Don’t send the ambulance to London without question.
w3w is a great tool, on top of other ways we use to communicate location. We’re not computers and there is no unambiguous way to communicate “I’m next to the tree” to another human. What might be interesting is response times and success rates for emergency services where w3w was used vs where it wasn’t. You can’t talk on false positives without looking at actual positives too. “Sometimes seatbelts fail, so we should stop using them and then come up with a better system”… come up with a better system, then we’ll talk about retrofitting.
Once w3w does the google maps bit and lets you save an area of map, the offline argument is also moot. It’s already almost moot as if I can get a phone signal I can probably get data… except of course places like snowdonia!
Do you know why ISBNs were designed specifically differently to other barcodes?
It's because you can transpose any two digits and the ISBN will still work. To counter the most simple human error seen when entering book numbers manually.
The same way that you know the code of any ISBN barcode is faulty if it doesn't total up to a multiple of 11 (where any X is taken as 10).
You can design systems to cope with errors. You're literally using probably a hundred different checksum and ECC and error detection systems just putting your post on this page, at every level from your keyboard up to the website's database's multiple storage drives.
If you want a resilient system, you can easily build one where it doesn't matter if people get 5 out of 10 numbers wrong, you would still know where they are. Literally you can choose the number of errors you want to correct and that'll determine the size of the final code and almost all such codes are not prohibitive. We literally use the exact same system to take to Voyager spacecraft where some countless thousands upon thousands of errors can occur in every megabyte of data transmitted and it'll still get through in an understandable way where you know it's correct.
W3W has *SO MANY* flaws that it really shouldn't be taken seriously.
Lat/Long or OS are long-standing, standardised, royalty-free and "just work", and you can do it in a foreign language (I know the German, French and Spanish for the numbers 0-9 without even trying, all I'd need is the compass directions and possible "minus").
But if we want a resilient system, random numbers or letters would actually work far better - even just a small bunch of alphabetical characters could do a better job than W3W, could be transmitted with the NATA phonetic alphabet or Morse Code, and just one or two extra characters would greatly enhance its resiliency in the cases of tranmission or transcription errors.
It's literally an afternoon jaunt for any mathematical student averse in Coding Theory, with a dash of geographical coordinate systems, to code up a solution to all the problems there.
But when Lat/Lon (international, standardised) and OS (UK only, standardised) are staring us in the face, there's really little need to do so. Transmit both. Transmit one and your approximate location. Or just read it twice.
But the problems with W3W go far beyond confusingly similar symbols (not something that Lat/Long or OS suffer from), and it really should be put to bed as anything but a gimmick to get your pizza.
and just one or two extra characters would greatly enhance its resiliency in the cases of tranmission or transcription errors.
Which is exactly what W3W does.
I don't have a dog in this race. I'm not a user or provider of emergency services. As Shannon pointed out, natural language has enormous redundancy, and as forensic analysts have pointed out, natural language is constantly self correcting (it's easy to just cherry-pick communications to get any meaning you want). W3W may be a good system or a bad system, but a feature is the enhanced resiliency in the case of transmission or transcription errors, due added redundancy and error checking.
The issue is that W3W is completely closed and proprietary. It is obvious that it isn't perfect and could be improved by some additional curating of the words and combinations allowed. Removing more ambiguities. Following up on any ambiguities that actually caused issues (for example, by spending money to reward emergency services who report all failures/confusion) and making sure they are removed from the system, etc.
Before anyone contracts with W3W for anything important (let alone life and death) they should (i) improve it, and (ii) more importantly, demonstrate their commitment to, and actual demonstration of, their improvement learning process.
As I understand it, that's a big part of the problem.
What3Words don't have the ability to change the words/fix failures - they'd need to release an entirely breaking change (so you then get into issues with people still using an old version etc).
A couple of years back, I was asked for a W3W by the emergency services. The location they got was incorrect (though that proved to be a failure of the app rather than issues communicating the three words), thankfully it didn't matter too much because the non-W3W location I'd given them was pretty unambiguous.
The OS Locate app is a far less problematic in my (limited) experience.
Exactly. Anyone with the forethought to install W3W as a location app could just as easily install anything capable of showing GPS based Lat/Long, or OS grid ref, or any local maping co-ordinate system in whatever country they happen to live or be in.
The ONLY thing W3W has going for it IMHO is the marketing and awareness raised in people who might not otherwise be aware of other, pre-existing solutions.
On the downside, it's a profit making "disrupter" that, as per most so-called "disrupters", isn't actually solving a problem, uses proprietary tech and doesn't appear to be amenable to criticism and the consequent fixes as evidenced by their "robust" defence when likely flaws are pointed out to them. If they were really all about doing good and not just making money, they would take constructive criticism on board and evaluate it, not just dismiss it as "unlikely" with no corroborating evidence. That's the typical response of a commercial business.
And as reported here fairly recently, so does Samsung and others are also looking at it. The iPhone version (as I understand it) only allows texts to the emergency services and requires a subscription after two (three?) years while the Samsung version allows texts to any number but only a certain amount each month and requires a subscription much earlier (from day one?). There are also Bluetooth-connected devices which can add similar functions to any smartphone - I looked one up for a relative fairly recently.
M.
"The iPhone version (as I understand it) only allows texts to the emergency services and requires a subscription after two (three?) years while the Samsung version allows texts to any number but only a certain amount each month and requires a subscription much earlier (from day one?)."
iPhone Satellite Emergency SOS also allows communication out to your emergency contacts listed in the Health app, including your location data and transcript of any information you exchange with the emergency services. It's read-only though; they can't communicate back.
And yes it's free for 2 years; they're mooting a subscription after that but details haven't yet been released.
I long ago decided that them finding me if I end up with a broken femur on a mountainside is too important to leave to a glass fondle-slab. So I spent a bit of money -- but less than a decent fondle-slab costs -- and now have a device that will not only send a GPS position to emergency services, but also sends a radio signal that passing aircraft can pick up. And I not need to pay a subscription fee or remember to charge it daily. And it floats. It just does one job, but it does this job fairly well.
My company gave me an Ocean Signal PLB1 when I was stationed in Africa; very small and light. Luckily never needed to put it to the test, but have no reason to suspect it wouldn't work as described.
Note that you MUST register it with your country's maritime rescue services (it works on land as well but all signals are routed through maritime channels), and the costs vary wildly from country to country. The UK allows you to register them indefinitely and for free, whereas a number of European countries charge anywhere from €17 to €87 per year to register. Some countries also make it difficult for you to cancel your registration; you get lots of threatening letters describing the dire consequences that will befall you if you transmit using the device after de-registration, as well as requiring you to request de-registration within a narrow time window in the last month before your registration expires. Failure to do this means you're automatically reregistered for the subsequent year, with costs, and zero option to cancel.
Speaking from personal experience. I'm not bitter. Well, not very bitter, anyway.
I forgot to de-register my boat radio when I sold the boat. Six months later the new owner "pressed the red button" prompting a call from the UK coastguard.
They were very nice about it and sorted the de-registration for me right then.
For those who don't know: The "red button" on the radio transmits a mayday along with the current location when pressed. It's connected to the GPS.
The standard IOS 16 Compass app doesn't display W3W location! Well, not the one on my iPhone.
Apple Maps also shows Lat/Long for current location, but does so in decimal. Whereas Compass displays it in degrees/minutes/etc. format.
Of course they can fix it: just use a totally new set of words with no collisions or homophones. And if they actually want to be global they should think about how it will work for someone who only speaks Fininsh, Mongolian, Japanese, Inuktitut or French. And hope they do not believe that there is an unambiguous 1:1 translation between languages.
Actually, digits are generally fairly unambiguous between languages, one could design a system where a location is determined by a sequence of digits, where you add digits to get more precision. We could all it "What12Digits" and sell it to emergency services globally!
When W3W tried to promote themselves by giving the location of the end of the queue to process to the queen lying in state 3 out of the first 10 locations were incorrect because the locations were passed on incorrectly. When that was ridiculed they had to take extra steps to get the W3W location encodings passed correctlly.
No, it does none of that stuff. It is an encoding system, not an error detecting, let alone an error correcting one. An error detection system would have additional tokens such that most incorrect combinations would simply be invalid. You could, for example, add in a fourth word which is based on the three preceding words and must be a certain one for each preceding group, thus a misheard combination would come up as rejected unless you happen to hit a workable combination by luck. Error correction would likely require even more words so the data could be sent twice. Their algorithm doesn't have that. I don't know why you said it. It is really making you sound like you either have no understanding what error correction is or how W3W works, and I don't mean from a technical perspective but the bit that's obvious even if your only experience of it was reading this article, where they point out that it doesn't do error correction.
From the rest of your statements, I'm going with both. For example, you say that "natural language has enormous redundancy". It doesn't, by the way. It has a bit of redundancy, but really not too much as you probably know if you've ever been misheard. However, this isn't natural language, it's words. You can probably guess the ends of my sentences if you've read the starting half because I'm making a single point. If, however, you read this clause and I invite you to guess the following ones: laminate energetic truckle facade, you don't have so much of a chance. All redundancy is lost even though the same tokens are used. You also cite some "forensic analysts" who say that "natural language is constantly self correcting". I don't even know what these analysts are trying to say, but I'm guessing it's not what you said.
Add a Luhn code to each W3W, like the last 2 numbers of your credit card?
It then becomes what three words and a 2 digit number....
Other question but I've not been arsed to check: do W3W actually represent an actual grid square that has been painstakingly mapped out or does it encode like gps points to avoid having a few hundred million records in their database?
natural language is constantly self correcting
But w3w doesn't seem to take advantage.
For an example of (mostly) well thought-out resilient natural language, one need look no further than the phonetic alphabet used over noisy links. Compare with the British version used during WWII and consider why Able Baker Charlie Dog was changed to Alpha Bravo Charlie Delta (etc.) Wikipedia
M.
If Inland Revenue calls taxpayers "customers", a 999 caller must be the same?
From there it's only a stones throw from invoicing emergency services with a 5 grand ambulance ride like The Land Of The Free?
Shades of the 0118 999 skit from the IT crowd comes to mind... nicer ambulances, better looking drivers, and mated to an IBM Support inspired service team "Emergency Services good morning, My name is Kevin and I'm here to help you. What is the nature of your emergency, but before we get started can you please give me your credit card number?"
"natural language has enormous redundancy"
Yes that's true, and certainly 3 words could work better than trying to describe a location. The problem is that if they did it properly, they should (a) never have allowed homophones* in their original word set and (b) excluded plurals altogether. And then use that knowledge to be able to decode for example "dogstart" to know that 'dogs' is invalid as a plural and therefore the correct phrase is "dog - start". English has 21 consonants, so even excluding pairs that sound the same (s/c, c/k, j/g etc), there are at least 15 quite distinct consonant sounds. And 5 clear vowel sounds. using simple short words with a maximum of 3 consonant and 2 vowel sounds leaves a corpus of 85000-ish words. It doesn't even matter if some are not real words, as long as their pronunciation and spelling is unambiguous.
Not saying that the idea per se is bad, because it isn't. Using map coordinate numbers requires saying a word for each number if saying it verbally, so requires a dozen words to get even a not-very-precise location. Using alphabetic characters is either keeping some ambiguity with letter sounds that sound the same, or thinking that everyone knows the alpha-bravo-charlie alphabet (and will end up with people in stressful situations trying to find a word that starts with p and settling on pterodactyl or philosophy)
* I appreciate that they HAVE attempted to remove homophones but apparently enough remain to create some confusion.
Ok I knew the bit about adding up to 11 but not the transpose but, that’s interesting.
Thing is, I’m not using checksums, my phone or computer is.
Possible !== Appropriate
The reason w3w exists is *because* it’s hard to relay numbers for GPS coordinates between people, especially under stress. The other systems are used by people in the military who are trained to deal with stress.
I appreciate the detailed and accurate argument but I am not convinced
If you look at W3W Company Accounts you will find that the company does not make money and never will. It will make money for a few company directors at the expense of of "investors" and crowd funders.
I expect they will finally go out of business some time next year.
"The notion it exists to solve a problem is nonsense because it does not actually solve the problem, but creates yet another flawed approach."
I considered the notion that US doesn't have an existing system other than Lat/Long, as I note that is sometimes the case when US "tech bros" come up with new standards or systems they plan to sell to the world when the rest of the world already has system in place. But, no, even that is no excuse as the US does have the United States National Grid. Mind you, that only became a standard in 2001, so maybe W3W wasn't aware of it?
> The reason w3w exists is *because* it’s hard to relay numbers for GPS coordinates between people
Let's run with that for a second.
You've called the emergency services and they ask for a W3W. If you don't have the app installed, they will generally SMS you a link that takes you to a page that can show your W3W.
At this point, we can surmise that you've got a data connection (otherwise you couldn't access W3W in the first place).
You then need to say "dog.start.fart" to the call handler, because that's easier to communicate than GPS coordinates.
The problem is, in this scenario, there should be no need to have you say anything **at all**. You've just clicked a link they sent you, and clearly have a data connection - the system could just as easily send them the W3W (or better, GPS coords).
In fact, there are better solutions already on the market that do *exactly that*. Having to say the words at all is because that's half of the W3W unique-selling-point.
Just to top it off:
- at different times, W3W have themselves advised that it should not be relied on in emergency situations
- They also insist that you should have read (and agreed) to their (many thousands of words long) terms and conditions before using it at all.
So, you shouldn't actually be using W3W when in situations involving serious stress in the first place.
They shouldn't be anywhere near anything that's life or safety critical.
If you call 999 directly the AML feature in your phone will transfer lat/long directly to emergency services. If you are in contact Mountain Rescue they will send you a text that sends you to a web page that will automatically receive your lat/long.
W3W uses word combinations that will not be intelligible if spoken under stress or with any kind of accent.
Found a burnt out car down a small lane in the middle of nowwhere, so reported it to the Police.
Shrunk it down to.
What road is it on?
Don't know, but I have the grid ref.
We can't use those?
This is OS grid ref, the standard way of IDing anywhere in the UK. What maps do you have then?
Gave up.
I wonder if it is still there or rusted away.
"If you muff the second or third decimal place of *any* lat/lon you can easily be several km out"
Assuming you meant taking 50.120 to 50.129, no, it wouldn't put you that far out, about 110 m. Even changing the second digit from 50.10 to 50.19 would be about 1.1 km, which is enough to cause a problem, but not quite as high as you state. This is why some code with error correction would be useful. However, the W3W system appears not to have any of that, just turning the numbers into words. If you know that you're going to hear a series of numbers, you have less possibility of mishearing than if you could hear any of ten thousand English words, especially if there is a bad connection or unfamiliar accent in play. On a bad phone connection, someone misheard me saying "I am tired" as "I was fired", so it's really not that hard to make a mistake. Their protocol doesn't have any correction mechanism, so you have to hope that they have placed any potentially misheard words very far apart. Being sent to Arizona instead of Britain is a giveaway, but being sent to somewhere twenty kilometers away might not be if the area is sparse enough.
With a single digit error in a lat/long together with a rough description of the location, e.g. Cumbria an error in a high order digit can be corrected, and error in a low order digit still gives a rough idea. With W3W the smallest error results in total loss of information.
But most importantly lat/long are used automatically all the time without problems. e.g. pin on a google map. AML in any emergency phone call, the mountain rescue system where you click on a link in a text to pass your location to a mountain rescue website.
W3W has been shown to have lots of similar sounding words in close proximity due to their algorithm using a much smaller dictionary in some areas.
Oh, and have you seen their accounts?
That problem can be solved with Four King Maps
Plot your position anywhere in our Great British Isles with four swear words. (It don't work in foreign parts).
https://www.fourkingmaps.co.uk/
e.g. scrotum.cocksucker.piersmorgan.arsepiece resolves to a spot just north of Dundee.
(Bit surprised El Reg was slow on the uptake of that one. I think they are loosing their touch.)
Oi, we wrote about that forking site in 2021, thankyouverymuch.
https://www.theregister.com/2021/08/14/sweary_four_word_map/
It's even in the more-stories list in the middle of the article. Consider yourself tsk'd.
C.
I can just imagine how this would go with the emergency services...
"Help, I've fallen on a mountainside and I think that my leg is broken"
"OK, can you give us your Four King Maps location and we'll send out mountain rescue to assist"
"It's scrotum..."
"OK"
"cocksucker..."
"Yes"
"piersmorgan"
"Don't you use language like that with us, sonny. You can die alone on that four king mountainside for all we care"
Yes, likewise a Devon based RAC patrol being assigned a job in Orkney, 750 miles and at least one ferry trip away :-)
The story doesn't explicitly state how this happened, but implies it was a Post Code error either in understanding a spoken character string or entering it into the system.
Devon based RAC patrol being assigned a job in Orkney
It's for a different reason, but around here (let's say Cardiff) systems often tell us our nearest "something" is on the other side of the Severn, perhaps in Weston Super Mare or Minehead. People over there probably get the same in reverse.
Relative was looking for a second-hand car recently, had a shortlist, asked for vehicles within 30 miles and probably about a half of the top search results were 30 miles in a straight line, significantly further by road due to the whole big-river-estuary-in-the-way thing.
M.
Yes, it's the same on almost all "distance" based searches on the web, whether that be Google, other search engines or, far more commonly, websites own searchs. Almost none of them take geography into account. I live by a river and the nearest crossings are 4 miles down river or 6 miles up river and there's a town centre just across the river. That makes searching for retail locations etc lots of fun since that is, by distance, the nearest town centre being only about half a mile from my house across a 400 metre wide river. Handy if I had a boat, I suppose :-)
I get the same on the Tarbat peninsula in the highlands... a lot of my "nearest" shops show up in searches as being in Elgin which by boat they probably are at around 25 miles but to actually drive there around all the water is nearer 75 miles and I'd have to go through Inverness on the way!
I thought it was in Warrington when you said that. It must be your accent. https://what3words.com/ship.ship.ship
Or did you mean Singapore: https://what3words.com/ship.sheep.ship
Oh sheet sheet sheet https://what3words.com/sheet.sheet.sheet
You could be absolutely anywhere
I hope you are not stuck somewhere that is steep steep steep https://what3words.com/steep.steep.steep
You could get quite mixed up with those similar sounding words
To me the solution is simple, a good, old fashioned checksum.
If the app displayed a checksum, the user could read out the three words and then the checksum number 0-99 and the operator would type it in to their app and immediately know if they've heard correctly. If it doesn't match, ask them to spell out the words.
The checksum wouldn't stop the standard 3 word combinations from working, just provide the additional confirmation that's needed when using the phone. I think users would find it easy to understand too.
The large company I worked for before retiring had an 8 digit employee number. Except it wasn't. It was a 7 digit number with a checksum as the 8th. Once I realised how it was calculated I used it in every program I wrote or worked on that needed the correct identification of an individual and logged the failures.
The log wasn't large but it was big enough for me to think that some people really needed a spellcheck for numbers as well as words.
Obligatory XKCD reference, latitude/longitude coordinates can scale from "It's on the Eurasian Continent" to "That atom, there":
https://xkcd.com/2170/
Pretty much every device that can do W3W can provide lat/lon coordinates, without license fees.
12 digits provides more than 3 times more precise measurement than W3W, 10 only slightly less accurate.
Either you're handing out raw floating point variables, or you've built a database to track individual atoms. In either case, please stop.
Example: Location: SD 75120 72690 / 54.1494253905477, -2.38241125754417
Pretty much every device that can do W3W can provide lat/lon coordinates, without license fees.
If you've ever tried to get a user to correctly read a simple error message to you over the phone, you'll know that expecting them to correctly read two 5-digit numbers is a lost cause.
W3W may have potential errors and ambiguities, but you've still got more chance of getting an accurate position than you have by asking the average user to give you lat/long.
It's more a problem for the excitable passerby: "There's someone knocked off his bike. Where? I think it's the B something, B1234 or may B4321? Mavis, what was that last village, Bumfuck under something? My position? 5123, sorry 26, then a dot, or maybe it's a comma, I haven't got my reading glasses on."
fair comment David, yes, it was a little unfair, but getting people to do stuff you need them to do should be in any public facing person's toolkit... [see hell desk/public facing anything]... I have to explain the phonetic alphabet to some people to get a valid postcode because their power is out and they desperately need to watch the next episode of "i'm a nobody what the hell am i on the telly for"
But why would you speak it? If you phone 999 the AML system will send your lat/long automatically. If you use google maps, you will drop a pin - which is just lat/long.
If spoken, you still need to spell out a random number of letters, because the words are too ambiguous.
Lat/long with a simple description still gives reasonable chance of correction with a 1 digit error. W3W loses all info with a 1 character error.
Lat/long will still be around in a year. Judging by W3W accounts they shouldn't have had their most recent rounds of crowdfunding to keep them afloat, but still sinking.
Yes, in most cases. However, if the connection was brief for some reason, for example the person is injured, and the person handling the emergency is far from the area, for example a state-wide emergency service because the local ones are overloaded, you may lose that information. This is why a word-based location system needs to be very careful about not putting things that sound similar anywhere close to each other, which might include keeping them out of the same large areas. If they don't have error correction, which they don't, it might be worth keeping all the possible clashes in other countries, or at least on different sides of a large country.
I agree, this example is it doing what it should.
The problem is though, that's what the system is *supposed* to do, not what it always does.
For example,
- you need an ambulance to likely.stage.sock
- The dispatcher's heard likely.stages.sock
They're within visible distance of one another, so the dispatcher may associate that with the approximate area you've given, but the ambulance is ending up on the other side of the river to you.
This isn't the only examples of this, especially once you start factoring in homophones and the like. If you look at Cybergibbon's post you'll see that, in built up areas, it's actually worryingly frequent
If What3Words worked the way that they claim it does, it might not be so problematic, the problem is that it doesn't, to the extent that even they've tripped over it in the past https://w3w.me.ss/even-w3w-gets-confused-by-plurals/
> To be confused with clip.apple.sleep, clip.apples.lip, clip apple.slip to name but a few.
There are many examples that are much more confusing. "slip" and "sleep" don't really sound very similar to me.
Having said that, I'm a Brit so perhaps adding some American drawl would make them sound more similar?
Take people with a lot of different accents and a bad phone line and the clashes will start to be noticeable. A lot of people speaking English will pronounce long E and short I differently. A person who has learned English recently and started with a different language may not. In fact, that vowel confusion is a stereotypical accent characteristic of a few languages that use more pure vowel sounds than English does, although its presence in practice varies a lot. Similarly, there are some languages that don't use the dental fricatives (the various sounds that th can make in English), and some people who have learned languages that lack them will do something else as they learn to make the same sounds. Two common methods effectively replace it with a t or s sound. Take a person like that and suddenly, words like think/sink or thank/tank become harder to tell apart, especially if you add in static on a phone or radio. This isn't only for people who recently learned English. For example, some Irish accents have a different pronunciation of th which is quite distinctive from t, most of the time, but close enough that I've seen a misunderstanding between an Irish person and a British person even though both were native English speakers.
This isn't only for people who recently learned English. For example, some Irish accents have a different pronunciation of th which is quite distinctive from t, most of the time, but close enough that I've seen a misunderstanding between an Irish person and a British person even though both were native English speakers.
I work with in a non-English speaking country with an Irish bioinformatician, when she first joined the Company and started talking about R she got nothing but blank looks because of her accent!
Almost a Guinness -->
"Having said that, I'm a Brit so perhaps adding some American drawl would make them sound more similar?"
Or just travelling a mile or two down the road, in the UK. Accents are blending more these days due to easy communications and national/international tv and radio etc, but not so very long ago, accents varied wildly from village to village and even different parts of town. As for the US, there a areas where they seem to go out of their way to never pronounce the letter H in many words and and other cases where they strongly emphasise the letter H. eg 'erbs and Ve-hicle. Then there's the UK divide over whether it's properly "an 'otel", "an hotel" or "a hotel". Lets not even go near the Scon/Scoane debate :-)
Or a few primarily British accents where or and aw sounds can sound the same, meaning that flaw and floor can't be distinguished by sound. I wonder whether W3W, which appears to be American (where telling those words apart is trivial) have that set in their list of homophones, extended to derivative words.
Or people pronouce vowls wrong.
Seen a bloke on TV who uses O for A, so plonts.
Then some TV historian, who on a quiz show sounds so weird you consider subtitles.
Best would be the west country who ER it.
Panders eats bamboo
Some northerners turn ERs into A so Panda and pander would both be panda.
Just program into the firmware the ability for the system software to send the GPS of the phone to a designated emergency service call centre via e-mail, text message, Facebook message, or by vocalising it out loud during a phone conversation. Press volume up, volume down, volume up, volume down (or the other way around) and it kicks in, offering the opportunity to request help.
The problem isn't just the sending, but the receiving too. You need your 999 service to be able to receive a text message and link it to the ongoing call. Like E911 or E112 do. Now you just need to make it a globally recognised and implemented standard, and hope that wherever you're calling from has enough connectivity to maintain a data connection as well as a phone call.
The bit you're suggesting is the easy bit, and has (rather unsurprisingly) already been done.
That already exists, it's called AML and supported on the vast majority of smartphones in use today. When you dial 999 the smartphone gets your location and sends a text message to the emergency services. The problem is that many 999 call centres have not integrated it with their systems.
Here's a real life example of W3W going wrong: Keswick Mountain Rescue Team
An 83 year old female collapsed in Crow Park. The informant gave the 999 call handler a What3Words location which was close to Hawes End. The team sent a Landrover first truck to the location only to find nobody. Further enquiries revealed the true location within walking distance of the base. Fortunately more team members had arrived at base and were able to respond quickly to this potentially serious medical incident. The casualty was assessed and stretchered back to base for further assessment and to await the arrival of an ambulance.
This is the second callout in 3 days (and there have been others) when the W3W location has been close enough to be believable but wrong enough to be useless. W3W should not be relied upon on its own. Always give a verbal description of where the casualty is and better still a grid reference from a map or use the app
The W3W combinations are designed to not produce ambiguous combinations *nearby*, which is great if the receiver of the w3w already knows roughly where the location is going to be, because it functions as a pseudo-checksum (e.g. if I'm giving my home address to you as a w3w, and you already know I live in Davenport in Stockport, for example, then there is likely only one possible match among the homonyms, etc, in Davenport); but the wider the area is, the more possible locations it could be. Seems like a design flaw.
I had to call 999 to request the local Cave Rescue to evacuate a companion with appendicitis. The police (who for historical reasons handle cave and mountain rescue calls) required a W3W location. I provided them with one (though I also have two OS grid ref apps on my smartphone). The operator then identified my as being at Ingleborough Show Cave, when I was actually at Gaping Gill, about a mile away with no vehicular access between the two.
P.S. See page 16 of Mountain Rescue Magazine
Why What3Words is not suitable for safety critical applications
Complaints were made, and that was the least of them. The mobile phone mast I connected to was in Lancashire, but I was in Yorkshire. It took 40 mins for the message to get passed from Lancs Police to Yorks and on to CRO. All that time I was standing away from the casualty on open moorland to get a signal (fortunately the weather was clement) before I was called back. Mobile connectivity is so poor in some mountainous areas that calls should never be terminated. The police refused to acknowledge any of this as a problem: all my complaints, to both police forces and the IOPC were brushed off.
"The police refused to acknowledge any of this as a problem: all my complaints, to both police forces and the IOPC were brushed off."
Too easy to do when the casualty survives. A coroner's verdict citing the problem might be trickier for them. As with so many other issues nobody wants to think ahead to anticipate the occurrence of a worst case.
Many, many years ago in 1985 I wrote a password generating algorithm for the company where I then worked. It took a 30 bit pseudorandom number, split that into three 10 bit fields and then spat out a three word password, with each word of the password taken from a list of 1024 words, all nouns, that had been carefully crosschecked for possible confusion by feeding them all through the "soundex" algorithm. Soundex is normally used to find words that sound alike, even if the writer uses the conventions of another language (e.g. "mare-duh" instead of "merde"). It was developed to help match people's names under variant spelling.
If as a fresh faced junior programmer I could manage to spot, and mitigate, the risk of homophones, what does that say about the quality of the thinking that went into What Three Words?
Admittedly, they have a much larger keyspace they're trying to include, which makes their challenge a bit harder. If they created a combination for every square on a Mercator map to 0.00001 degree, they'd need five words from a set of 1024 to include them all, which they evidently thought was too long. Of course, they could try improving this by tossing out the squares that are the open ocean, antarctica, etc, but they decided not to. The problem is that, having found this problem, they decided to allow plenty of confusable pairs, but to make their algorithm try to keep apart the pairs they thought of. That's a rather weak way to deal with that problem.
It's fairly clear that they didn't even do that, they just assumed the homophones and singular/plural forms would be a long way apart.
And they keep claiming this despite being presented with multiple cases pairs being within 10-20 miles, including in their own advertising.
It would have been relatively simple to (eg) ensure homophones and singular/plural forms were diametrically opposed on the globe.
Heck, even simple duplication would have worked better - it's generally fairly easy to determine if you're in the Pacific Ocean or Europe.
I'm sorry, but that isn't a valid Four King Maps reference. However, bum.fuck.willy.shit describes a location half way between King's Lynn and Norwich
Depends how old you are as it's been around since 1980. But it's mainly an amateur radio thing, often used in contests for targets like how many major squares you can make contact with in 24 hours. It's usually used with four or six character precision (about 100km and 10km respectively) but the standard goes to eight characters or about 1km.
So while nowhere near close enough for emergency use, it has the benefit like Lat/Long and the OS Grid of being monotonic, so comparing two locations it doesn't take much mental arithmetic to see which direction one is from the other, whether it's a long or short distance and whether you're getting closer or further. Unlike w3w.
Yours, from IO91ul...
https://en.wikipedia.org/wiki/Maidenhead_Locator_System
I suspect it could be extended to provide a better resolution. And from a UK point-of-view, it would be easy to teach people "you live in India-Oscar 92" or "Juliet Oscar XX" (for Norfolk, Suffolk, Essex, and Kent), and then teach them the smaller numbers and letters. A suitable phone app could convert their Long/Lat to an accurate Maidenhead Locator, and spell it out in the NATO phonetics for easy transmission. The call-handlers are all trained and would (should) be able to pick-out the word, even with noise in the background.
The Maidenhead Locator can be transferred under some pretty awful radio conditions. I took part in the RAF Air Cadet Exercise Blue Ham (Exercise Hermia) this June 2023, and the cadet stations from the RAFAC, Army Cadets, Sea Cadets, and Combined Cadet Force managed to exchange information through serious QRN (thunderstorms raging across Europe) and major QSB (very broken sky conditions from mutiple X-ray flares thrown at us by the sun). It takes time and many repeats, but when you know you are listening for IO or JO, the rest is fairly easy.
73 from somewhere in IO92ub
> It has received endorsements in the UK from the Automobile Association (AA)
Just try calling out the AA if you don't have a smart phone or don't want to install their app. I gave them the exact address and postcode of the place I was broken down (within greater London), and the road name and number. But because I couldn't give them the three magic words, they sent the tow truck to a place two miles away.
They seem to be completely dependant on W3W and they can't cope with anything else.
I didn't have much problem a year or two ago. No mention of W3W, however I took a great deal of care to ensure that my location was described exactly.
Me: Clapham Station, North Yorkshire
...
Me: Yes, that one, not the one in London. Yes, it does have a Lancaster Postcode.
I am disabled and suffering from Diabetes.
I was on a trip to collect something I had bought on eBay, when I had a puncture. Fitted spare wheel and continued. On the way back, only about ten miles from home, had another puncture, but of course, no spare now. Phoned the AA and gave them my location, right beside a phone mast on a bridge over an A road. "we'll be with you in ten minutes".
To cut a very long story short, three and a half hours later the van pulled up - on the hard shoulder of the A road fifty feet below where I was on the bridge. The driver then took a further quarter of an hour to navigate the back roads to reach me, where he put on his universal spare wheel and followed me the ten miles home. As I hadn't had any food for over six hours, I was feeling extremely wobbly by the time we got there.
Complained to the AA, but their excuse was that it was a Bank Holiday Saturday, and the changeover weekend between two different Half Term weeks, so the traffic was heavier than usual. If they knew that, why did they keep on telling me they would be with me in ten minutes?
When you call the emergency number, your phone transmits your location at the same time, and it's free, and the mobile network provides a location too (presumably based on cell tower triangulation)
https://eena.org/our-work/eena-special-focus/advanced-mobile-location/
> AML stands for Advanced Mobile Location. In the event of an emergency call, an AML-enabled smartphone (all Android and iOS devices worldwide) automatically sends accurate location information of the caller to the emergency services. This information is derived from the location data of the phone (GNSS, Wifi).
> AML is not an app; it does not require any action from the caller. AML is simply a protocol to transport the data (using SMS and/or HTTPS) from the smartphone to the emergency call centre. AML is – of course – free of charge. Emergency services are then able to receive this information in all the countries that have deployed AML.
https://eena.org/knowledge-hub/documents/aml-in-the-united-kingdom/
> Handset locations obtained through the AML functionality are be compared to the location provided by mobile networks (using cell coverage information), using an algorithm that analyses factors such as time of positioning and the separation of the two locations.
> Once the mobile handset knows its location it is sent to BT using a simple, already available, Short Message Service (SMS) based protocol (which gives 160 characters of data).
If you don't know where you are and your phone isn't able to use GPS to locate you, W3W is going to be just as bad as AML.
people seem to like the idea, several people have extolled its virtues to me but none had investigated beyond the advertisers hype
w3w is just an attempt to monetise latitude and longitude.
in UK OS NGR: SK123854 is a 10metre square and these options are usually good, not perfect but neither is W3W
LatLon: 53.365, -1.817 Postcode: S33 7ZA Location name: Edale Station
Google maps "share my location" works well, sends a link to a map with a point marker, you need mobile signal but without that summoning assistance becomes a bigger challenge anyway.
AML seems to be widely implemented and SMS can get sent even where mobile signal is too weak for voice.
Am I the only one that finds such an application utterly pointless? If you already HAVE the lat, long GPS co-ordinates on your phone, make those available as part of the 999 call and/or other tools as needed.
I'll grant not everyone can read out numbers larger than their number of fingers; but if you get the application right (e.g. your phone dialler) there should be no user input needed.
The slight problem with lat/long, from the perspective of the untrained person who is possibly using them for the first time in what's likely to be a bit of a panic, is the multiple different formats and reference frames that lat/long can be conveyed in.
For the sake of argument let's assume the lat/long you want to convey is referenced to the WGS84 geoid.
Do you see it as:
Decimal degrees,
Degrees and decimal minutes,
or Degress, minutes and seconds
and then - whether or not negative numbers are used to convey locations south of the equator and west of the Greenwich meridian.
Is it interpreted by the receiver as another format?
If lat/long positions is read out by the user without understanding, it can be interpreted as another very easily.
The error in translation may be relatively insignificant (~500-1000m depending on circumstances), or may be highly error prone, or could result in an invalid coordinate).
This is why, I presume, UK Fire/Police/Ambulance have used OS eastings and northings (easily translatable to OS grid references) for years. Other than the numbers, there's nothing to misinterpret.
Speaking from experience after 22 years in mountain rescue.
Overall, I like the idea of a word-based method of giving locations. This is based on my experience of people who cannot remember, nor are able to read out, any number with more than three digits, combined with those who cannot accurately write down/type any number with more than one digit. I absolutely agree that things like AML are probably the best way to give locations, but only as long as there is no transcription needed anywhere in the process - sometimes even 10 metres out is too much. However, any word-based solution has to be done right if it is to be used for emergency situations, and that isn't easy due to things like accents etc.
Until such a system comes along, though, many of the problems can be dealt with by a) getting people to enunciate the dots - dogs DOT toe DOT dearth is clearly different from dog DOT stowed DOT earth - and b) asking for spellings - W3E words aren't long, and can be spelled quite quickly. Again, there are accent problems if the speller doesn't know the phonetic alphabet, but it reduces the error-space quite significantly.
Every situation where W3W is possible to use is machine to machine.
I click "share location" in Google Maps or FB Messenger etc etc, you click the link.
AML automatically shares my location with the emergency services when I dial 999, 112 or 911.
I paste lat/long into an SMS.
At no point is a human actually reading it out, they all get a pin directly on the map.
And if they don't use (or have access to) the same proprietary map system as me, they can copypaste the lat/long from any of them to any other. Something that is deliberately impossible with W3W
Sneaky of W3W to use extremely convoluted numerals to buttress its claim that words are easier to remember than numbers. In practice, nobody quotes LATLONGs. They text it and the recipient just clicks the link. Ergo, there's no need to speak out, much less, remember numbers like 40.712772, -74.006058. Besides, there are alternative systems like LinCode, which use a sequence of pure numerals like mobile numbers or credit card numbers, without any decimals or plus / minus signs, and they're way easier to speak out than W3Ws like mine: bristle.slam.parsnip.
I remember one particular callout somewhere in the southwest of England a couple of years ago. The customer decided to helpfully give our control what three words for the site as they said the postcode would take you to the wrong place. Control helpfully recorded this information and passed it on to the engineer. The engineer dutifully loaded up what three words and entered the three words and got somewhere in Asia - Kazakhstan IIRC. They tried the post code which got them to street level. They then called the customer on arrival and found the entrance to the customer site just round the corner on a side road. As such the postal address was perfectly fine to locate the address. If you don't do field work or indeed deliveries you might be surprised how often you have to call the customer when you arrive as the site entrance isn't immediately obvious from the street. It is however vanishingly rare for the postcode to be several thousand miles out.
The engineer asked about the three words and found that one word was missing it's final letter S and that was what resulted in a massive error. And right there is the problem with what three words. It's very easy to hear similar words incorrectly and the only way to ensure errors like this don't happen is to go through the spelling of all three words using the phonetic alphabet. You can get 11 metre accuracy using lat long at 4 decimal places. So generally around 12 digits. That's quicker than going through three words using the phonetic alphabet.
In other words the idea that what three words is easier than lat long or other map references is bogus. If you clearly hear the words then yes great. However how sure can you be that you've heard the words correctly? In a life and death situation you're going to want to be 100% sure and the only way to be sure is to fall back on the phonetic alphabet.
I'll stick with lat long or map references thanks all the same
With landlines, the emergency service can look you up in a database and instantly know the exact address of the caller. No-one was complaining about lack of privacy. Mind you, your address wasn't broadcast with your call. It had to be cross-referenced via a resource only available to the emergency services.
Now for the modern equivalent: Surely smartphones can embed the GPS data in the call metadata except... privacy. So how about an encrypted system where the GPS data does not exist in the 'clear' but as a key or token that can be looked up on a database or tool that only the emergency services have access to? The encryption could be periodically refreshed so an old static database does not provide a lookup. Yes, something could get hacked yada yada. But that's always a risk anyway.
From the article:
"If W3W is widely adopted by emergency services it must be subject to rigorous evaluation," he explained in a preprint paper titled: "A Critical Analysis of the What3Words Geocoding Algorithm."
The Register contacted Arthur to talk about his findings but he declined, stating that he hopes the paper will be published in an academic journal and that he would rather wait until the peer review process is complete before discussing the work.
Hmm, well, if he didn't want to discuss it prior to publication, why put it up on arxiv?
I'm a little confused as to what the point of the discussion is anyway. So far as I can tell, W3W are saying that for locations near, for example, Crewe there are unlikely to be clashes / confusions and you'd double check anyway (i.e. correct for gross, wrong-continent errors). Whereas everyone else (such as this preprint?) seems to be considering the matter globally? It seems fairly reasonable that, in most applications, you'd anyway double check that when someone says "over.ages.they" they do actually mean "West Street in Crewe", after which you can be confident that they're outside the Crewe Heritage Centre.
I can understand problems occuring when a phone's location services are a bit unsure and are just reporting position guestimated from what cell towers it can see. You can see that on Google Maps sometimes, when it paints an enormous circle to say "you're somewhere in here, maybe".
That's not to say that W3W is the best way of verbally conveying accurate location. As the article mentions, the OS Grid is pretty much ideal for this kind of application here in the UK.
Or another way is to simply assume that whatever emergency service you're calling can guess your lat / long to within a degree anyway, just through which cell tower you're using, and you just quote 8 digits that are the numbers after the decimal point to 4 digits (which is pretty accurate). Those could even be sent down the phone call as DTMF sounds. (patent pending)
'I can understand problems occuring when a phone's location services are a bit unsure and are just reporting position guestimated from what cell towers it can see. You can see that on Google Maps sometimes, when it paints an enormous circle to say "you're somewhere in here, maybe".'
It's exactly ^this^ which is the problem.
w3w is marketed as being a solution to being found an emergency, pinpointing the location to 3m.
So, perhaps folk download the app and try it at home. Mere seconds pass by, and they get a satellite view of their back garden. Wow, they say. Most impressive!
They don't appreciate they're in range of multiple cell sites, multiple nearby WiFi networks, and have a good strong data signal to help with A-GPS and hotspot/cell tower databases. All these helpers mean their phone can get a location, *fast*.
Then weeks later, out in the sticks, perhaps under some tree cover, in the range of only one cell tower, with no nearby WiFi, no/poor data signal, and in a panic not really knowing exactly where they are.... they open the app and rhyme off the first three words they see to the 999 call handler. They don't notice the massive blue circle. They don't appreciate they're relying on GPS acquisition alone which could take many seconds, perhaps even a minute or two, to get an accurate location. But it's too late. They've passed on those three words and they're in the system now.
Things have got slightly better. w3w produced a script for call handlers to follow; but fundamentally their system is still subject to spelling and transcription errors.
Speaking from experience.
If you need fast and easy transfer of location information, learn the NATO alphabet (Alpha, Bravo, Charlie, etc) and use something like Open Location Code (aka + codes, supported by google maps, and should be in Apple Maps too iirc). Faster, easier, less confusion possible.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
