[Talk-ca] What do I poutine the name tag of a road with a suffix?

[Minh Nguyen]

Vào lúc 17:06 2022-12-11, Hoser AB đã viết:
> I also sympathize with Minh's point about TTS systems not being able to reliably disambiguate. That said, I do find it funny that Minh muses:
> 
>>I can't help but wonder if some of the motivation to abbreviate the street names comes from how folks would like to see openstreetmap-carto display them. Maybe that isn't "lying to the renderer" per se, but it is ironic, considering that other tile layers like Mapbox Streets and OpenMapTiles would abbreviate these quadrant suffixes anyways. Surely the tradeoff of having to tag name:pronunciation would annoy most mappers; that key is only designed for overrides for unavoidable edge cases. 
> 
> Well Minh, if it's for the expressed purpose of making it easier for TTS engines to synthesize spoken directions, isn't changing the name in and of itself "tagging for the renderer"? You brought up an example from Lethbridge, "Sometimes [a TTS engine] get[s] lucky with partially abbreviated text, for example with '2 Avenue S' ('two avenue south')". (https://www.openstreetmap.org/way/296323533) I had to laugh, I couldn't help it: it's not "two avenue", it's SECOND avenue! :D So as much as the "tradeoff of having to tag name:pronunciation would annoy most mappers" it seems we're already doing this for your benefit anyway: we're just fudging the name tag to make it (sort of) work for you. I appreciate Minh that you deal with TTS and wayfinding for a living and have worked on as you put "special-cased" things to work better in Canada, but by the same logic we ought to be writing out "one hundred and sixty-second avenue southeast" instead of "162 Avenue SE" (https://www.openstreetmap.org/way/459124027). 

Glad you noticed! This is another thing that trips up anyone using OSM 
data for text-to-speech in Alberta. You've probably encountered it if 
you've ever used OSM for navigation there. It would also affect other 
places like New York City and Washington, D.C., but there the name=* 
tags use ordinal numbers despite the cardinal numbers on the street name 
signs.

In fact, the <say-as interpret-as="address"> extension does optimize 
numbers to some extent: "1020" goes from "one thousand twenty" to 
"ten-twenty", which is much more comprehensible. For more basic TTS 
engines, some resort to more clever hacks. [1]

But what if these TTS engines did add some smarts to insert ordinals 
automatically? Or what if, as a hack, the application sticks an ordinal 
into the SSML before sending it off to the TTS engine after some delay?

"Arriving at 1020 Century Drive, on the right"
-> "arriving at ten-twenty century drive, on the right"
"Arriving at 1020th Century Drive, on the right"
-> "arriving at one thousand twentieth century drive, on the right"

"In 200 metres, turn left onto Highway 63"
-> "in two hundred meters, turn left onto highway sixty-three"
"In 200th metres, turn left onto Highway 63rd"
-> "in two hundredth meters, turn left onto highway sixty-third"

Of course you could imagine the extent to which a navigation 
application, after going out of its way to determine that the maneuver 
is in Alberta, could add double and triple exceptions to fine-tune its 
output there. I'm not holding my breath though.

Realistically, guidance instructions in Alberta will always be a little 
rough around the edges, and I suppose that's a necessary tradeoff for 
only tagging streets with their names as written. When you introduce 
your acquaintances to OSM and get questions about why the instructions 
sound funny, at least we'll have had this discussion that you can link 
them to.

> Brian happened to provide examples from the US of A to illustrate that 
> in said country they've made the choice to "unabbreviate" where 
> possible, and that's fine and dandy; it doesn't matter much to me what 
> the nomenclature is in Washington, District of Columbia, United States 
> of America (although I have been to the White House and happen to know 
> that the street signage quite consistently uses abbreviations, i.e. 
> "Pennsylvania Ave NW" :P). I do find it amusing that his other example 
> came from Salt Lake City; a cursory perusal of their streets using 
> (*eeek!*) Google Streetview shows that by and large the street signage 
> is usually unabbreviated as-is. For example the street signage by the 
> Mexican Consulate at the corner of 700 South and 200 East does in fact 
> say "700 South" and "200 East" (verbatim; majuscules, miniscules and 
> all). Of course no local mapper in Salt Lake objects to this: that's how 
> they named their streets.

The authorities in most of the U.S. don't use sheet metal as the primary 
medium of passing local ordinances on street naming. The sign is just a 
best-effort attempt to communicate the designation to the public. When 
it comes to abbreviations or the lack thereof, the government trusts 
residents to use common sense and abbreviate or expand as the occasion 
deems necessary.

For example, my city has a practice of alternating the spelling of its 
numbered streets in numerals and words from block to block. [2] If you 
go down a non-numbered street, you'll see signs for "First", "2nd St", 
"Third Street", and so on, up to "Twentyfourth St" and "Thirty-third 
St". [2] If you go down 10th Street and look to the side, you'll see 
signs for "10th" and "Tenth" alternating every other block. These are 
all technically "North # Street" or "N # St", but they don't post the 
cardinal direction because it would be repetitive.

In another city I lived in, two signs at the same intersection would say 
"E 8th St" and "Eighth Street East", and oh by the way there's a giant 
sculpture of just an "8" to make the city's preference clear. [3] Some 
road names identify the road's endpoints; the signs alternatively put a 
space or hyphen in the middle. Other roads are maintained jointly by 
local and state authorities who, for political reasons, reverse the 
street name and engage in sort of a turf war with their signs. [4]

Whenever Americans point out how complicated things can be over here, 
some Europeans like to say "Then map the signs!" Sometimes that's all I 
can do. [5] Not that any data consumer actually uses this information, 
but sometimes it's helpful for heading off confusion among mappers. 
Typically, when mapping specific traffic signs, you'd need to know the 
standard sign code. For example, a street blade sign in Ontario would be 
tagged traffic_sign=CA:ON:G216.

Unfortunately, the Alberta Highway Guide and Information Sign Manual 
gives no code for rural or urban road name signs. [6] If anyone knows of 
an appropriate code for these signs, perhaps the Canadian road sign 
tagging guide could be expanded with that information. [7]

[1] https://www.mapzen.com/blog/voice-guidance/
[2] 
https://lists.openstreetmap.org/pipermail/tagging/2021-January/058958.html
[3] https://www.mapillary.com/app/?pKey=1133045673830592
[4] https://www.openstreetmap.org/changeset/68488944
[5] https://overpass-turbo.eu/s/1oWs
[6] 
<http://www.transportation.alberta.ca/Content/docType233/Production/Guide_info.pdf#page=58>
[7] https://wiki.openstreetmap.org/wiki/Canada/Road_signs

-- 
minh at nguyen.cincinnati.oh.us