the essence of place
“To catalog is not merely to ascertain, as it appears at first glance, but to appropriate.”
— Roland Barthes, “The Plates of the Encyclopedia”
I am an avid user of Google Maps. Not just for navigation, I find simply browsing Google Maps an inherently enjoyable activity. In an increasingly digitized world, somehow I think I can understand a place simply by looking closely enough at it on Google Maps. I’m fascinated by label placements. I see them as almost implicit markers of a place’s definitive location; the one spot you absolutely must visit to understand the essence of a place. City labels especially are fascinating on Google Maps. From far away, cities are rendered simply as a dot with some text next to it, but if you zoom in to a precise level, you’ll see the label is actually affixed to a very specific spot:
Here is the label for New York City. The text “New York” is located directly above city hall in the Financial District. I would not describe this as the defining location of New York City by any means. Why not Times Square or something? Though I can respect placement on top of a seat of government as a rule, and after all the Financial District is the historical city center. But don’t even get me started on the label for Brooklyn:
The more I think about this label placement the more it pisses me off. Perhaps I could forgive the label for being so far from the geographic center of Brooklyn if the intersection of Brooklyn Ave and Atlantic Ave were at all remarkable. But no it’s just a random intersection along the A/C line with a 7/11 that the Google Maps label for Brooklyn just so happens to be positioned directly on top of. Why not just put it in the actual center of the borough? What purpose does this serve? Who or what algorithm is responsible? I descended a rabbit hole.
Cartography is a very old discipline which, like many things, underwent significant change with the advent of the digital era. Manually placing labels on maps, such that they don’t overlap with each other and are aesthetically pleasing, was a tedious task, often taking as much as half of the total time of map creation (1). As modern cartography developed over the course of the 20th century, generally well-accepted labeling practices emerged, such as those published by Swiss cartographer Eduard Imhof (2). According to Imhof, there are three types of labeling problems:
Point-feature labeling i.e. how do you label things which are just dots on a map, such as cities.
Linear-feature labeling, like roads and rivers.
Areal feature labeling, for features which cover large areas, such as entire countries or oceans.
Most cartographic labeling algorithm literature focuses specifically on the point-feature labeling problem. Algorithmic point feature labeling, such that every point on a map is given an aesthetically pleasing label that doesn’t overlap with another label, is actually a fascinating combinatorial optimization problem. Algorithmic approaches use Imhof’s guidelines, like that the ideal position for a point-feature label is to the top right of a point, to assign every potential position for a label a numerical score, with an example ranking scheme shown below:
For every label on the map, its score is combined into an objective function, such that minimizing the objective function would produce an ideal map labeling. Mathematically, this problem is quite difficult, and finding the true global minimum of the objective function is intractable for all but trivially simple maps. A significant body of research focuses on applying heuristic methods which don’t find a global solution but are good enough for practical purposes (3).
Areal features don’t generally involve the same level of combinatorial complexity, and as such don’t receive the same attention in literature. Most algorithmic solutions simply try to put areal labels somewhere towards the center of the shape it’s supposed to represent, often modeled as the centroid of the polygon, but the flexibility of labeling an area rather than being attached to a single point makes overlapping less of a concern.
But where do you draw the line between a point and areal feature? As Imhof remarks, the difference between a label as an areal feature and a point feature is largely dependent on scale. For a static map at a fixed zoom level, such as what most map labeling algorithms in literature were designed for, the difference can be easily defined. But Google Maps is not a static map; it’s a dynamic map that lets you zoom into any level you please. At what threshold does a city go from just a dot on a map to a shape with area? City labels on Google Maps represent a nebulous gray area where because of the dynamic nature of the map, they have to act as both at different zoom levels. Realistically what’s the alternative? That at a certain zoom level, the New York label suddenly moves to become an areal label and fill more space, rather than remaining fixed on a single point? This would probably be a bit jarring from a user perspective. But it’s the point-area duality of city labels that enables a Google employee to implicitly pick the precise spot which defines New York City.
All things considered, this is probably a relatively benign example of how Google Maps shapes our perceptions of reality. But map-making label placement choices like these, even if algorithmically assisted, should not be thought of as rigorous scientific decisions, but as a reflection of the cartographer or company which placed it. With the emergence of modern Geographic Information Systems and computer-assisted tools, cartography increasingly became perceived as a strictly objective, scientific discipline. This growing conception of cartography was challenged by J. B. Harley in his essay “Deconstructing the Map” (4), employing Jacques Derrida’s postmodernist approach of deconstruction to understand maps not merely as objective scientific representations of reality, but ultimately acts of rhetoric:
“The steps in making a map—selection, omission, simplification, classification, the creation of hierarchies, and 'symbolization'—are all inherently rhetorical. In their intentions as much as in their applications they signify subjective human purposes rather than reciprocating the workings of some ‘fundamental law of cartographic generalisation.’”
There is no such thing as an unbiased map, because the act of map-making itself involves the interpretation of geographic reality into some form. Because of this, Harley argues that cartography is also an exercise of power by those who make it, whether used intentionally or not:
“Power comes from the map and it traverses the way maps are made. The key to this internal power is thus cartographic process… To catalogue the world is to appropriate it, so that all these technical processes represent acts of control over its image which extend beyond the professed uses of cartography.”
Google Maps, and the exact coordinates it chooses to place city labels, is an expression of power by Google itself. In this light, OpenStreetMap is a fascinating alternative to examine. It is essentially the Wikipedia of maps; an open collaborative project attempting to provide a freely usable geographic database of the world. On OpenStreetMap, cartographic debates over label placement take place out in the open on the internet rather than privately amongst Google cartographers. The city point-area label dichotomy appears in OpenStreetMap city labeling guidelines. Places in OpenStreetMap, meaning any city, town, village, or even notable unpopulated area, can be mapped either as full areas with a defined polygonal border, or as a single node with no border. Because of the frequent uncertainty of exact city borders, the OpenStreetMap wiki recommends modeling cities only as nodes, but nevertheless many users choose to model them as areas instead. Area-defined cities cause enough problems in the default OpenStreetMap renderer that cities mapped as areas are intentionally not rendered with a label at all, leading to the necessity of mapping a city’s borders with an area while including a separate node to place the label.
Digging into changelogs for the New York City node led me to a shocking discovery: Since 2007, the label for New York was not placed over city hall, but instead over a random spot in the East Village just off the intersection of 2nd Ave and 11th St. There it remained for over a decade, except for being moved slightly further down 11th St by user starwars425 in August 2018. But it wasn’t until April 23, 2019 when OpenStreetMap user LukeyBear audaciously moved the city labels for Los Angeles, Las Vegas and its suburb Henderson, Philadelphia, Washington D.C., and New York to directly over their respective city halls all in one fell swoop that the New York label arrived at its present location:
A more pernicious example of cartographic decisions shaping perceptions of reality is neighborhood labeling. Neighborhoods are essential components to the fabric of a city, such that the socioeconomic ramifications of living in a particular neighborhood are well-studied, and neighborhoods are often treated as a fundamental unit of urban planning studies (5, 6). Despite this, there’s no universally agreed upon definition of a neighborhood, because in the end they are vague sociological constructs which mean different things to different people. Using a data visualization technique designed to communicate the uncertainty of an AI model’s evaluation of cancerous prostate tissue, Adam Pearce wonderfully captured the ambiguity of New York neighborhood boundaries using drawings submitted by 15,000 readers of DNAInfoNYC asked to draw the boundaries of their neighborhood. People generally agree upon the center of a neighborhood, but not much beyond that.
The same node-area duality manifests in neighborhood labeling as well, with the OpenStreetMap wiki advising only to map neighborhoods as areas if there is a well-defined legal or administrative boundary associated with it, but recommending “where the borders are fluid or there is no broad agreement on where the boundaries are located, then it is best to use a node positioned in the centre of the area.” But Google Maps tends to have fewer scruples about defining exact neighborhood borders, often to the dismay of people who live there. Google frequently uses data from unofficial sources not intended as an official reference, leading to situations where Google legitimizes realtor invented neighborhood rebrandings, or where a typo on a Detroit neighborhood map made by an urban planner as a side project leads to Google Maps almost single handedly renaming the neighborhood “Fiskhorn” to “Fishkorn.” Or when Google Maps suddenly relabeled the historically black neighborhood of Fruit Belt in Buffalo, New York to “Medical Park”, a move seen by many residents as a sign of gentrification. Perhaps these are unintentional mistakes, and Google is largely operating in good faith, but Google Maps’ omnipresence as a mapping tool means these labeling decisions have significant ramifications for how these neighborhoods are perceived.
Do these controversies sound familiar? They remind me an awful lot of the conversation surrounding the Spotify hyperpop playlist. I see the act of grouping the disparate cultural threads and internet scenes under the “hyperpop” label and Google Maps neighborhood labeling as precisely the same “non-neutral act of referring.” Genres and neighborhoods are both inherently vague sociological constructs, given exact definitions by tech companies in the form of playlists and cartographic labels respectively. I think it’s instructive to view the hyperpop playlist, as opposed to a major label dominated hip hop playlist like RapCaviar, as a manifestation of what sociologist Sarah Thornton calls “subcultural capital” (7). Thornton developed the notion of subcultural capital in her book Club Cultures: Music, Media, and Subcultural Capital as an extension of French sociologist Pierre Bourdieu’s concept of cultural capital to explain the social dynamics of British youth club movements in the 90s, describing the status gained from one’s cultural knowledge or attunement to an “underground” movement. Many of the same observations Thornton made can be applied to the complex tangle of internet movements and subcultures sometimes referred to as hyperpop. She emphasizes the role niche journalism plays in the development of subcultures:
“They categorize social groups, arrange sounds, itemize attire and label everything. They baptize scenes and generate the self-consciousness required to maintain cultural distinctions. They give definition to vague cultural formations, pull together and reify the disparate materials which become subcultural homologies. The music and style press are crucial to our conception of British youth; they do not just cover subcultures, they help construct them.” (p. 151).
In the development of hyperpop as a subcultural identity, this was a role largely fulfilled by the Spotify playlist, rather than the press. But a playlist makes no room for context, flattening the expansive network of collectives, sounds, and histories which make internet music so fascinating into a decontextualized, but expertly curated, amalgamation of songs. And ultimately driving playlist creation decisions, and the subcultural capital derived thereof, is the Echo Nest data platform, specifically whatever clustering algorithm is putting a box around patterns of sonic characteristics, listening behavior, and the web-crawled internet conversations surrounding them and elevating it to the level of genre. Echo Nest developers may point out the inherent uncertainties surrounding their approach, that their model perhaps shouldn’t be considered dogmatic ground truth for what is and is not a genre, but you wouldn’t know that looking at the playlist.
Cook, A. C., &; Jones, C. B. (1990). A Prolog rule-based system for cartographic name placement. Computer Graphics Forum, 9(2), 109–126. https://doi.org/10.1111/j.1467-8659.1990.tb00384.x
Eduard Imhof (1975) Positioning Names on Maps, The American Cartographer, 2:2, 128-144, DOI:10.1559/152304075784313304
Christensen, Jon, Joe Marks, and Stuart Shieber. 1992. Labeling Point Features on Maps and Diagrams. Harvard Computer Science Group Technical Report TR-25-92
Harley, J. B. (1989). Deconstructing the map. Cartographica: The International Journal for Geographic Information and Geovisualization, 26(2), 1–20. https://doi.org/10.3138/e635-7827-1757-9t53
Durlauf, S. N. (2004). Neighborhood Effects. Handbook of Regional and Urban Economics, 2173–2242. https://doi.org/10.1016/s1574-0080(04)80007-5
Matthews, S. A. (2008). The salience of neighborhood. American Journal of Preventive Medicine, 34(3), 257–259. https://doi.org/10.1016/j.amepre.2007.12.001
Thornton, S. (1996). Club Cultures: Music, Media and Subcultural Capital. Wesleyan University Press.
harrd post omg :0 the scale of maps thing will mess u up i remember travelling without my phone once and i looked at google maps before i left to get an idea of where to go and i was like "sick this town only has two roads i just have to turn left at the second one to get to where im staying" but as soon as i got off my bus i realized the town was 50x bigger than i thought it was and i just hadnt zoomed in on google maps to see all the roads and i was suddenly hopelessly lost