Proquints: Identifiers That Are Readable, Spellable, and Pronounceable (2009) | Hacker News Reader

Proquints: Identifiers That Are Readable, Spellable, and Pronounceable (2009)

arxiv.org

82 pointsingve25 comments

English pronunciation mentioned together with words “unique” and “unambiguous”―is that a joke? Oh yeah, no way that people pronounce ‘u’ and ‘a’ similarly.

“There are 6 unambiguous 1-letter vowels”―since when are vowels in English unambiguous? English writing should properly be classified as logographic, since you can't know how a word is pronounced from its spelling, or vice-versa.

Has the author ever communicated his own middle and last names in a phone call? A big benefit of numeric identifiers in the first place is that it's easier to dictate them without error. Even in my native phonemic language, with words or letters it's too often back to the Folks' Phonetic Alphabet: “No, D as in Dmitry.”

> Oh yeah, no way that people pronounce ‘u’ and ‘a’ similarly.

Take their example "gutuk-bisog." The second 'u' could easily have a schwa sound.

Now imagine someone pronounces that string over the phone. The recipient then asks the speaker if that vowel is one of [eauo]. Now the speaker's brain uncontrollably fits whatever wrong letter the recipient utters into the slot to realize the ambiguity.

Now the recipient had better hope the speaker just hard-coded "second vowel is 'u'" into their memory. Because there's no other information to resolve the ambiguity.

Yet how would the speaker have known ahead of time to do that?

People using this system would have to learn to be more precise with their pronunciation and say a true /u/ instead of letting it become a shwa.

This could still work in situations with some kind of error correction. Like "read your user ID" where gutuk-bisog is a valid user id but (e.g.) gutak-bisog is unlikely to be.

The Upspin project uses these to make private key recovery easier. I love the idea.

However, the choice of letters only seems unambiguous to western speakers. In Arabic, for example, P and B sound the same, and so do V and F for most native speakers.

Yeah, uniquely pronounceable subset of English has no vowels in it, so it's not pronounceable at all :)

Just use a sane phonetic language.

The strings only have to be unique and pronounceable to be valuable, they don't have to be uniquely pronounceable.

When scanning a list of records by eye looking for an entry or entries, it's helpful for the identifiers to be pronounceable because that allows normal people to hold more in their short term memory than they could if the strings were all gibberish and each entry ended up consuming multiple short term memory slots in the brain. Sure, there is a tiny chance of collision but that chance is still there when scanning by eye and trying to find a GUID that starts with the first few letters of the GUID you're looking for. At that point you verify you actually have the one you want. And yes, there are lots of occassions where people look at identifiers, that's why pronounceable identifiers have value.

I'm baffled as to what you imagine the use-case to be. Is that how e.g. IP addresses are used, in your opinion―by finding them in an onscreen list? Or SS numbers? Or credit card numbers?

Replacing hexadecimal stack traces?

As others mentioned, this is the least feasible use-case exactly because the time window is short and the return for the introduction of the new scheme is dubious.

Perhaps something like the Japanese Kana could be used instead?

https://en.wikipedia.org/wiki/Kana

See also the Bubble Babble encoding (2000) [1] that is essentially the same idea. Note that they are only useful as private identifiers because as public identifiers you can't avoid them spelling profanity; many geocoding systems entirely forwent vowels for that reason [2].

[1] http://wiki.yak.net/589

[2] https://github.com/google/open-location-code/wiki/Evaluation...

I (re)discovered a similar scheme about 7 years ago for encoding numbers based on the observation that 20 consonants + 5 vowels offers essentially the same information density as base 10 numbers. Excluded 'c' because it seemed the most redundant (with 'k' and 's'). I've found it particularly useful for generating random temporary passwords that can be communicated verbally rather than needing to be emailed.

https://github.com/femto113/node-pronounceable

Interesting, how the example ip address conversions and up sounding vaguely Turkish or Arabic...

To me, they all sound like the names of lost cities from H. P. Lovecraft.

Not since the wretched dark sun last rose over Lusab-Babad, or the accursed sorcerer-priests of Haguz-Biram last spilled human blood on their altars, had such indescribable evil lurked in the shadows as now did amongst the filth-ridden canals and dank buttresses of Budov-Kuras, Budov-Kuras which squatted upon the forbidden ruins of Mudof-Sakat, even its name blotted from memory by the horror of its crimes.

I recognized Hebrew in it, especially "todah" (thank you in Hebrew).

I think urbit uses something like this for it's address space.

I find Urbit's significantly more pronouncable. I wonder if it was inspired by proquint.

Urbit's "phonetic base" is much more pronouncable because it always allocate carefully chosen three letters per octet [1].

[1] https://stackoverflow.com/a/38175707/225272

Another idea: find 256 different syllables and use that to encode 8-bit numbers. This way we also can avoid profanity (except for the rare case where a string of multiple harmless syllables expresses a profanity).

or find 65,536 words and use them to encode 16-bit numbers, then you could leverage your existing word associations to memorize

NOTE: 2009 should be in the title.

>Debuggers could input and output memory addresses as proquints as an alternative to hex.

...but why? In debuggers, you're not concerned with memorising the address of an object as much as you are concerned with what the hell is mangling said object. There's no "value-add" for proquints, here. If you were live-debugging in a session with someone else, sure, but how often are you generally doing that?

>Network tools such as browsers, ping, netstat, traceroute, etc. could input and output proquints as an alternative to dotted quads.

Again, what is the value-add? Since I've known 4-octet IP addresses and numbers my whole life, it would take more effort to translate from proquint, to understand the output on the screen. Adding a translation layer in a cut-over fashion instead of a phased-in fashion just seems like wholly unnecessary overhead, only intended to further justify the proquint, yeah?

Yes, that debugger example really confused me and made me wonder why he added it. Anyone who has actually done live-debugging sessions will know that hex is perfectly pronouncible, and furthermore is a direct mapping to binary.

True, even in post-mortem debugging (assuming I have the objects from the heap), it's not like I'm going to exclude the addresses of objects in my write-ups (or as they're exposed via commands, such as !sosex.mdso[0]).

In native, if you're on the stack (e.g.: not using pointers), then addresses don't mean much of anything and if you're using pointers, memory optimisation means the address could mean feck-all after a cycle or two (assuming you've looking at an iDNA/TTT[1]).

So, to agree with you, who are agreeing with me: It is quite odd that this would be a principal argument for proquints.

[0] - https://github.com/lowleveldesign/debug-recipes/blob/master/...

[1] - https://docs.microsoft.com/en-us/windows-hardware/drivers/de...

"The suggested optional magic number prefix to a sequence of proquints is "0q-"."

That is interesting - when I designed "Oh By Codes"[1] I chose "0x" as the prefix to identify that sequence.

My entire life, every time I have seen a hex number I have pronounced the prefix in my head as "Oh By ...." - hence the name.

[1] https://0x.co

What could possibly go wrong?https://news.ycombinator.com/item?id=10409837

Reply on news.ycombinator.com