Can You Know A Book Better Without Reading It?
Data Visualization as a way of deriving meaning
When I was young, I was a voracious reader. Fantasy and science fiction, essays, role-playing game manuals, philosophy, history, news, culture, you name it. I was in the honors program at a Jesuit college where I majored in ethics and ancient history. Before I washed out of my PhD program without finishing my dissertation, I stacked another few dozen books and who knows how many papers on top of that.
And then it stopped. I stopped reading. I hardly read at all now. Not in a dystopian sense — so much of our media is text — but no novels, no books, no essays. When I do read it’s to re-read an old favorite that’s part nostalgia and part communion. I haven’t replaced it with audiobooks — which I find a poor substitute simply because the voice of the reader is not my own interior voice so it feels less like reading and more like just another product to consume. The reason I loved to read was not because I got some college or cultural credit, but because it afforded me the chance to open a dialogue with myself on the subject of what I read. For me, an audiobook is a one-way transmission.
When I do read some research paper or some article on climate change, I do so in the debased way I learned to devour texts in grad school — a very practical and not at all meaningful way of approaching reading. Now, I’ve replaced reading with experimental data visualization. I extract the data from the media, where possible, and explore it by visualizing its contents. This treats written and other works as systems for playful examination. I’ve done this not just with books but with television shows, like Archer, or games that I’ve never played (and will never play), like Pokemon.
So, when I saw on Instagram the cover of Sanzo Wada’s A Dictionary of Color Combinations, I was interested in it, but had no desire to read it. Back when I did read, I read a lot about color theory, and while there’s a lot to know about color, a work like this dictionary isn’t going to give you more color knowledge as much as it will prove a lens into the thinking of the designer who wrote it. Before the Internet was drowned in tools for choosing palettes, books like these were useful for inspiration. That’s no longer the case — now you can find a thousand palettes or make your own from any photograph.
Fortunately for me, Dain M. Blodorn Kim transcribed the contents of the dictionary into structured data as part of a project to create an interactive version of the dictionary. And so, while I have never looked inside Sanzo Wada’s book, I saw that it was a collection of palettes, a common resource and, to me, almost like a rite of passage for a world-class designer — like how every architect designs a chair. I found the simple list of palettes to be rather boring and wanted to see the colors as a system, and so plotted a network of the palettes, connecting each palette to another palette if they shared colors. The results were similar to what you saw in the image at the beginning, a cluttered mess unimaginatively referred to as a “hairball” by network visualization practitioners.
So I did what anyone does when they are trying to derive meaning from a work like this, I processed it. We do this all the time, we read an essay by Mark Twain and take from it on one day an expression of sublime language or on another a critique of colonialism. But, when I did that with books, it’s always been intuitive and vague; when I read systems via data visualization, I’m forced to be explicit in how I choose to process it. I didn’t choose to visualize the data rather than read the book because it was easier or more effective, I did it because it was the only way I found interesting to engage with the book. Reading it didn’t interest me.
I pared down the network to only show the connections between palettes that shared more than two colors in common. When I did, a Sanzo Wada system offered itself. The nodes, in this case, are palettes, displayed graphically as rings of the colors that define them, looking much like gobstoppers:
The edges (the lines connecting the palettes) are rendered in the same way with the shared colors overlapping on the line. If you squint, you can see them. Some features immediately jumped out at me: most palettes were functionally orphans (disconnected from any other palette), and there were interesting structures like the palettes that share two shades of violet.
A more effective way to visualize network data is to separate out the components and place all the orphans in their own group (an orphanage, if you will) because the position of the nodes in many traditional network layouts is meaningless — only the connections matter. This view shows in more striking relief the patterns of shared color between palettes and the number of palettes that are not connected by two or more colors.
The use of so many high key colors usually begs for a black background, but Sanzo Wada used black in many of his palettes and so it was a dead end. I suppose this is exploratory data analysis but without any explicit goal. It’s not really meant to understand the dataset, per se, but rather to experience and derive some meaning from it. It allows me to engage with the idea of color, the idea of declared relationships between abstract subjects and the idea of knowing a person who I know literally nothing about. I could have looked Sanzo Wada up on Wikipedia and given a short gloss on him but why?
And just as I had no interest in knowing who Sanzo Wada was, I didn’t care about the orphan palettes. But I also didn’t like the way I was rendering edges. Like most data visualization choices, my earlier method of rendering edges was not based on effectiveness but ease. It’s much easier to draw a line over another line than it is to offset them. Fortunately, I long ago wrote a library for rendering all sorts of network edges, including parallel edges, so with a little effort, I drew the shared colors side-by-side. It grew much more readable and especially benefited the colors that were not high key, which made me see so many more of the subdued hues, greys, neutrals, and blacks that were more common than I realized.
Because of the subject, I thought it might be better rendered using a hand-drawn style. These are colors, after all, in a book from before computer-assisted graphic design, and a thick inky stroke seemed suitable.
But even though I never opened the book, just based on its cover, I felt like this approach didn’t fit. The clean lines of the cover screamed Bauhaus to me, and, surprising as it may be to a modern designer who needs only use a fill tool, earlier designers were actually capable of making very neat filled shapes and even straight lines, and I thought that it seemed more suitable that way.
The gobstoppers had to go. That was easy enough, replacing the custom node glyph with a different kind: a simple circle-pack made up of the colors in the palette. As with adjusting the edge graphics, this immediately made the system much more readable, bringing forward the same desaturated or neutral colors that were helped by the parallel edges.
At this point, the system takes on a much more network-centric view. There are some very interesting central nodes in the largest component, like the red-yellow-green-black palette and the pale-blue-pale violet-pale yellow-charcoal palette.
The new circle-packed nodes deal with sketchy rendering better, but the color effect of sketchy rendering is especially inappropriate for this subject. It makes the colors seem lighter and can overlap with the edges in unintended ways. This is less important than that it does not align with my vision of a book I will never read.
I then added some labels — the ID values of the palettes which seem to make no sense whatsoever. Palettes that share many colors in common seem to appear in very different places. I could have replaced the labels with something procedural, based on the names of the colors or some other gimmick, but that seemed useless. So, instead, I labeled the features I found interesting.
Labels in network visualization are a fraught subject and it seems most practitioners can only turn them off or turn them on and can only label the individual nodes and edges themselves. That’s far less useful than labeling constrained regions that indicate the patterns and meaning you feel you’ve discovered in the text.
For me, the most prominent was the group of fully-connected nodes (called a clique in network science) of seven palettes that all shared the same two shades of violet. They varied in size and additional color, with only two palettes sharing a third peachy color in common. In contrast, the shared use of lime green created only a tenuous connection between palettes that otherwise had little in common.
The only thing left that still interested me was to revisit the orphans where I could see prominently the single-color palettes that make this work even more mysterious. What is a single color palette? How was Sanzo Wada using them? So many of the palettes we see today are thematically defined by their color space: desaturated, high key or the always popular pastels. But these palettes seem to cover the gamut from primary colors to many near greys and everything in between. The orphan palettes were more sedate than the connected palettes, which seemed to be dominated by cleaner colors.
In retrospect, I could have “read” this work differently. Instead of a simple threshold of two shared colors, I could have used a percent of the palette, which would have brought the single-color palettes and more two-color palettes into the network structures. Or I could have opened the book, if only on some Google Books scan or perhaps even looked for it in a library.
The death of the author is historic fact, this is the death of the book itself.
But who am I kidding, I don’t read. And, in a way, I’ve experienced A Dictionary of Color Combinations like I never could have simply by reading it. The death of the author is a historic fact, this is the death of the book itself. I would argue, in fact, that I experienced it better by processing it than having simply read it. And, of course, the idea of experiencing a thing “better” is absurd, because I did not do this to improve myself or develop a skill or achieve some practical gain, I did it just to do it. I did it for the same reason I used to read: for pleasure and perhaps to find meaning. I achieved both.
What if the more you grow your graphicacy the more your literacy withers?
We often pose data visualization literacy (or graphicacy) as an additive thing, one that will make you even more literate than you were before. But what if it is subtractive? What if the more you grow your graphicacy the more your literacy withers? As I mentioned in an essay about creating that Archer visualization I did with Mara Averick, I found the experience of exploring the system of Archer in many ways more fulfilling than just watching the show. I have no desire to play Pokemon or watch Pokemon cartoons or Pokemon movies but I find analyzing Pokemon data to be engrossing. I cannot imagine that I would have had as meaningful an experience reading Sanzo Wada’s dictionary as I did exploring its system in this way.
I wonder what that means.