Polly pronounces “S” in ref as “South”

[freenerd]

In Saxony, Germany there are roads like S 81. Wikipedia has a list and here are some on the map.

Right now these are extended to South 81 instead of staying S 81 (via this code).

This is definitely a fringe problem (only 5000 km of roads in Germany have this abbreviation) but highlights issues with using abbreviation expansion of one place in another.

/cc @1ec5 @bsudekum

[1ec5]

Do you have a screenshot of the problem? Abbreviations.plist is used for abbreviating, not for expanding abbreviations, and only on the road name (after any ref has been stripped away). The expected behavior is that “South” in a name tag becomes “S”, but “S” in a ref tag should remain unchanged. This functionality is used in the turn banner, where street names are sometimes abbreviated to fit in the allotted space.

On the other hand, there is a geographical issue with Shields.plist, where CA may denote California or Cantabría, or NH New Hampshire or National Highway (e.g., in India), or CO Colorado or “county”. We can’t add support for German shields to the turn banner, because we can’t tell whether to use a German, Austrian, Australian, or Chinese shield for A. Unfortunately, there isn’t a better way to distinguish shields internationally until either Project-OSRM/osrm-backend#482 or Project-OSRM/osrm-backend#333 is implemented.

[freenerd]

I'm pretty sure the voice instruction said "South 94"

http://map.project-osrm.org/?z=17&center=51.380530%2C14.079157&loc=51.379536%2C14.076104&loc=51.378826%2C14.082263&hl=en&alt=0

[1ec5]

That application uses Amazon Polly. It’s likely that Polly interprets “S” as “South” in this context. We can avoid that and similar issues by surrounding the ref in a <say-as> tag that spells out the ref.

[1ec5]

We’re already marking up the ref to be interpreted as an address. I wonder if there’s a more appropriate interpretation here.

/cc @bsudekum

[1ec5]

On the other hand, there is a geographical issue with Shields.plist

→ #334

[1ec5]

@bhousel pointed out that, because of how many county roads in New Jersey are currently tagged ref=(510) instead of ref=CR 510 – long story – Polly pronounces them e.g. “open-parenthesis 5-1-0 close-parenthesis”. Those ref tags will be fixed in short order, but some roads in other regions have legitimate parentheses, like the A1(M) motorway in England.

[1ec5]

The “S” as “South” case was fixed in #509. Not sure whether “A1(M)” still gets pronounced as “a one open-parenthesis em close-parenthesis”.

[1ec5]

#552 should’ve fixed the parenthesis case as well.