The words of all literary works which were presented at the festival form the basis of a computationally generated key visual used across different media and formats. Thousands of words in four languages have been placed on a map using Yahoo’s geocoding service. The resulting points are then reconnected to form a graphical map of a fictional poetic landscape.
We have created two versions of the visual. One version is based on a vertical Buckminster Fuller’s Dymaxion map for use in the poster, a second version is based on the Albers Equal-Area projection for all horizontal formats, such as the website or projections during the festival. In addition to these two formats, the computational process allowed us to generate maps for each single author, with every map being an abstract visual identity of its submitted works.
The commission was meant to continue a much respected series of more than 10 years of work in which the basic principle remains the same: all works which are presented at the festival must be represented in the key visual in some form. The literary works provide the ›data‹ for an abstract illustration and thus the framework for a procedural design process. In its 14th iteration in June 2013, Poetry On The Road presented works of 25 authors from 15 countries in 4 languages.
Given that natural language processing, visualisation and experimentation with literature is one of our many interests, we have decided to approach the project from a data visualization perspective.
We were especially interested in how word occurrances vary between the works, a method that was inspired by stylometric research of one of our collaborators in the TransVis project: Jan Rybicki. Together with his team, Jan basically develops statistical methods that allow the identification of authors in arbitrary collections of books. His research on author fingerprints provided the starting point to our approach.
A second interest during the design process originated from our fascination with maps. We were fascinated by the idea to create a spacial relationship between the words through databases of placenames. Every time we search Google Maps for a place, such a relationship is established. But what happens if the search term is a single word in a literary work as opposed to the name of a business or known location?
To follow this idea we have first lemmatised all words to normalise their countless variations. All lemmata were then geocoded using the Yahoo geocoding API. Interestingly, the resulting map showed a nice distribution across the world but with clear foci in Europe and the Americas. Whether this is the result of the bias in the placenames database or the actual correlation between the language and areas of its cultural influence remains open. But the resulting distribution was a good basis for the next steps.
In the visual design phase, we have moved away from familiar map displays due to the fact that one of the many requirements for the key visual was a clearly abstract form. Based on the exchange with the client, we have moved away from our experiments on the use of patterns and maps and looked into other ways of creating shapes from a collection of dots on a map. We finally decided on the creation of contours shapes, which are a known visual element in topographical maps. This enabled us to keep a map-like formal language while not displaying an actual map, but rather the shapes resulting from isoline-based terrain generation process. We based the actual implementation on onformative’s excellent example of how to use a blob detection algorithm to create contour lines. With some minor tweaks and manual corrections we were then able to output the terrain graphics to PDF using Processing. In a final step we have added multiple dashed lines on top of the terrain as a visual reference to mapping grids.