Landmark queries

A landmark query names a specific, usually unique, place. The user doesn't want the nearest instance of a category — they want the Eiffel Tower, the Golden Gate Bridge, the Empire State Building. The geocoder's job is to recognize the landmark name and resolve it to a single coordinate or small candidate set.

What landmark queries look like

Query	Type
Eiffel Tower	Named landmark — globally unique
Golden Gate Bridge	Named landmark — unique, but spansspanA contiguous range of characters or tokens in the input string, tagged with an address component type (street, locality, postcode, etc.). Parsed addresses are represented as collections of spans, possibly nested in a tree. a distance
Empire State Building	Named landmark — unique, has a streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. address
Taj Mahal	Named landmark — unique, well-known
Sydney Opera House	Named landmark — unique, well-known
the White House	Named landmark — unique, politically sensitive
Wrigley Field	Named venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. — unique, has a streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. address
Grand Central Terminal	Named transit + landmark — unique, transit hub
Central Park	Named area — large, multiple entrances
The High Line	Named linear featurefeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them. — spansspanA contiguous range of characters or tokens in the input string, tagged with an address component type (street, locality, postcode, etc.). Parsed addresses are represented as collections of spans, possibly nested in a tree. a distance
The Las Vegas Strip	Named area — not a single point, not a polygon
Stonehenge	Named landmark — remote, well-known
Mount Everest	Natural featurefeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them. — summit vs base camp are different locations

Landmark queries differ from address queries in that the user knows the name of the place but not (or doesn't care about) its address. The geocoder must map the name to a coordinate without the structured components (street, city, state) that address queries provide.

Landmark taxonomy

Named structures

Buildings, bridges, towers, monuments: Eiffel Tower, Empire State Building, Big Ben, Colosseum, CN Tower. These are the easiest landmarks to geocode — they typically exist in gazetteersgazetteerA geographical index that maps place names and postcodes to real-world coordinates. Mailwoman uses a custom-built Who's On First (WOF) SQLite database as its gazetteer — the 'atlas' half of the grammar/atlas architecture. (WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations., OSMOpenStreetMap (OSM). A community-curated global map database (ODbL-licensed) with addr:* tagged features and place hierarchies. A secondary corpus source and a source of street names., GeoNamesGeoNamesA free global gazetteer combining administrative, postal, and POI data across 200+ countries. Supplements Who's On First for postcode centroids and places where WOF has gaps.) with a single name and coordinate.

The complication: some named structures have the same name in multiple cities. There are replica Eiffel Towers in Las Vegas, Paris (Texas), and Tokyo. The user searching for "Eiffel Tower" from Paris, France wants the real one. The user in Las Vegas wants the replica. The geocoder needs a prominence signalprominence signalA ranking input that favours the better-known place among same-named candidates — the original Eiffel Tower in Paris over a replica in Las Vegas. Closely related to the importance score. — the original Eiffel Tower is more prominent than the replica, and users searching without a location constraint probably mean the prominent one.

Named natural features

Mountains, rivers, lakes, forests: Mount Everest, the Amazon, Lake Tahoe, Central Park. These are areas or linear featuresfeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them., not points. A geocode for "Central Park" could be the park's centroid, the main entrance, or the Park's visitor center — all different coordinates. A geocode for "the Amazon" is meaningless as a point — the Amazon rainforest spansspanA contiguous range of characters or tokens in the input string, tagged with an address component type (street, locality, postcode, etc.). Parsed addresses are represented as collections of spans, possibly nested in a tree. 5.5 million square kilometers across 9 countries.

The geocoder should return the featurefeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them.'s centroid for map display and the most relevant entrance or access point for navigation. Which one to return depends on the use case, and the geocoder cannot know the use case from the query alone.

Named venues

Stadiums, theaters, museums, arenasarenaA standardized test set probing one capability: libpostal (clean canonical), perturb (noisy and degraded), postal (edge formats). Each arena answers a different question about where rule vs neural wins.: Wrigley Field, Sydney Opera House, the Louvre, Madison Square Garden. These are easier — they are buildings with streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. addresses, and gazetteersgazetteerA geographical index that maps place names and postcodes to real-world coordinates. Mailwoman uses a custom-built Who's On First (WOF) SQLite database as its gazetteer — the 'atlas' half of the grammar/atlas architecture. typically have them. But a venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. may have multiple entrances (main entrance, box office, VIP entrance, loading dock), and the geocoder's single coordinate is an implicit choice about which entrance matters.

Named areas

Neighborhoods, districts, informal regionsregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.: SoHo, Shibuya, the French Quarter, Silicon Valley. These are areas, not points, with fuzzy boundaries that change over time and vary by who you ask. A geocode for "SoHo" places a pin at an arbitrary point within a neighborhood whose boundaries are contested.

WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. modelsneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' neighborhoods as neighbourhood placetypesplacetypeThe Who's On First hierarchical classification of places: planet → continent → country → region → county → locality → neighbourhood. The resolver uses placetype to rank candidates — an exact locality match outranks a county-level match. with a representative coordinate and a parent chainparent chainThe sequence of administrative parents above a place — Springfield → Sangamon County → Illinois → United States. Used to check the geographic coherence of a parse. (Soho → Manhattan → New York City → New York → United States). The coordinate is approximate. The boundary is not modeled. This is acceptable for most use cases — "find SoHo on a map" — but not for use cases that need boundary-level precisionprecisionOf the spans the model labeled as a given tag, the fraction it got right. High precision means few false positives. Paired with recall to compute F1. (delivery zones, real estate searches, neighborhood statistics).

Named linear features

Trails, parkways, pedestrian paths: the High Line, the Appalachian Trail, Route 66, the Las Vegas Strip. These are paths, not points. A geocode at the midpoint of the High Line places the pin in the middle of the park — useful for map display, useless for navigation (the nearest entrance might be a mile away). Linear featuresfeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them. need a point-set representationhidden stateThe model's internal vector for a token after the encoder has mixed in surrounding context. The contextualized representation the classifier head reads to assign a label. (the path as a polyline) and a way to answer "where is the nearest access point?"

How traditional geocoders handle landmark queries

Google's Places API handles landmark queries through its place database. Landmarks are a type in the Places taxonomy. Google's results include a location (coordinate), viewport (bounding box for area landmarks), and types (what kind of landmark). Google handles disambiguation through user context (the user searching from Paris gets the real Eiffel Tower; the user in Las Vegas gets the replica).

Nominatim handles landmarks through OSMOpenStreetMap (OSM). A community-curated global map database (ODbL-licensed) with addr:* tagged features and place hierarchies. A secondary corpus source and a source of street names.'s tagging. Named buildings (building=yes + name=Eiffel Tower), monuments (historic=monument), parks (leisure=park), and natural featuresfeatureAn input signal a model conditions on. Beyond the raw tokens, Mailwoman feeds soft features — gazetteer-membership channels and the postcode anchor — that inform predictions without overriding them. (natural=peak) all appear in OSMOpenStreetMap (OSM). A community-curated global map database (ODbL-licensed) with addr:* tagged features and place hierarchies. A secondary corpus source and a source of street names.. Nominatim's search matches the landmark name against OSMOpenStreetMap (OSM). A community-curated global map database (ODbL-licensed) with addr:* tagged features and place hierarchies. A secondary corpus source and a source of street names.'s name tags and returns the best match. The limitation is OSMOpenStreetMap (OSM). A community-curated global map database (ODbL-licensed) with addr:* tagged features and place hierarchies. A secondary corpus source and a source of street names. coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. — well-known landmarks in well-mapped areas are covered; lesser-known landmarks in unmapped areas are not.

PeliasPeliasAn open-source geocoder, Mailwoman's spiritual predecessor. indexes WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. venuesvenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. and landmarks. WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. has good coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. of globally prominent landmarks (Eiffel Tower, Taj Mahal) but spotty coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. of local landmarks (the neighborhood park, the local statue). PeliasPeliasAn open-source geocoder, Mailwoman's spiritual predecessor.'s search matches the landmark name against WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations.'s name fields and returns the best placetypeplacetypeThe Who's On First hierarchical classification of places: planet → continent → country → region → county → locality → neighbourhood. The resolver uses placetype to rank candidates — an exact locality match outranks a county-level match. match.

What Mailwoman does today

Mailwoman's resolverresolverThe component that converts parsed address components (locality, region, postcode) into coordinates by looking them up in the gazetteer. The resolver ranks candidates by name match, population, and proximity, and returns the best-matching place with its centroid or polygon. indexes WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. administrative places plus a limited set of WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. venuesvenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label.. Well-known landmarks that exist in WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. as venue or neighbourhood records are resolvable. Lesser-known landmarks are not.

The parser handles landmark queries as venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. queries: a query with no address-format tokenstokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. (no numbers, no street suffixesstreet affixA modifier on a street name indicating type or direction — Street, Avenue, rue, Calle, N, East. Mailwoman tags these as street_prefix / street_suffix, recognized via a morphology FST., no stateregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. abbreviations) and a known venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. name is classified as kind=venue and routed to the venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. resolverresolverThe component that converts parsed address components (locality, region, postcode) into coordinates by looking them up in the gazetteer. The resolver ranks candidates by name match, population, and proximity, and returns the best-matching place with its centroid or polygon.. The F1 ≈ 0.39 figure once cited here was measured in v0.4.0, three modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' generations back, and no longer reflects the current classifier — a venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label.-fragment retrain is spec'd but not yet run. Treat the venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. classifier's quality as unmeasured until that retrain lands rather than trusting either the old number or an assumed improvement.

Landmark queries are partially in scope, and partially overlap with the newer category-query path: a landmark that also matches a POIpoint of interest (POI). A named place that is not strictly an address — landmark, transit stop, venue, amenity, or franchise. Mailwoman tags these as venue and resolves them through the gazetteer.-taxonomy category (a hospital, a named park) can resolve through poi.db the way amenity queries do. Bare landmark names that aren't categories ("Eiffel Tower") still depend on the venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. classifier and WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. records, where the limits are data coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. (WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations.'s venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. catalogue is not comprehensive) and the unretrained classifier described above.

See also

• Amenity queries — the generic-category version
• Transit queries — transit stations as landmarks
• Exotic POI overview — the series index
• Resolver and Who's On First — the gazetteergazetteerA geographical index that maps place names and postcodes to real-world coordinates. Mailwoman uses a custom-built Who's On First (WOF) SQLite database as its gazetteer — the 'atlas' half of the grammar/atlas architecture. that stores landmarks