Gary Hill: Writing Corpora
Gary Hill: Writing Corpora
Gary Hill: Writing Corpora
Gary Hill: Writing Corpora
Gary Hill: Writing Corpora
Gary Hill: Writing Corpora
Gary Hill: Writing Corpora

Gary Hill: Writing Corpora

February 2012

Artist: Gary Hill. Production: Reilly Donovan. Exhibition: Active Presence - Action, Object and Public. Venue: Museo MARCO, Vigo, Galicia (Spain). Curators: Sergio Edelsztein and Kathleen Forde

installation interactive kinect performance processing projection sound

When first getting hired on at Süperfad I created a set of custom software for media artist Gary Hill’s 2012 work Writing Corpora. The new work was created for the international group exhibition Active Presence: Action, Object and Public, which debuted at the Museo MARCO in Vigo, Galacia (Spain) during February of this year. The exhibition, curated by Sergio Edelsztein and Kathleen Forde, focused on artists whose practice contains both performance and installation elements.

The overall gist of Writing Corpora, without paraphrasing the artist too much, is physical gestures triggering text, video and audio relating to idiomatic phrases that refer to the human body (i.e. “put your foot in your mouth”), in both English and, in this version, Galician (the native tongue in Vigo). The piece is a continuation of the artist’s conceptual work focused on the convergence of body and language utilizing our current era’s ever evolving new technologies for self-expression.

For this project Süperfad provided technical support and code development, creating a software framework that allowed the artist to play back media elements throughout the gallery space with no physical controller other than the performers movements. The result is a full-featured toolkit that can trigger any audio and/or video media with practically any body pose or gesture.


We created two systems of custom software for Writing Corpora. The first was a fluid-dynamics “touch-floor”, an alphabet soup of letters that congeal into legible phrases and words when certain regions of the floor projection are stepped on, which is also an interactive element that remains for current and future gallery attendees. When a phrase formed in the text fluid it also simultaneously played an audio clip on overhead speakers of the same idiom being spoken in Galician if the text was in English and vice-versa.

The other software, used only during the performance by the artist, tracked skeleton data from a Kinect depth camera for real-time control of audio/visual elements with physical body gestures and movements. This tracked the physical distance between almost every possible combination of skeletal joints (elbow to head, foot to torso, knee to neck … you name it) and played back specific audio and/or video clips on one of three projectors in the room while a forth displayed this tangle of gesture data overlaid on the user’s tracked skeleton.


I was contacted by Gary’s studio assistant Reilly Donovan in late December to write custom code to fulfill the artist’s concept as they were hitting barriers with pre-built software. During this time Süperfad was bringing me in as their lead creative developer, and this project came with me. Süperfad founder and director Will Hyde turned out to be a fan of Mr. Hill and not only agreed to send me to Spain to help install the work but also sent along art director Loren Judah to assist and document the process. Primarily we were all excited about a collaboration in the realm of “pure art” guided by Gary Hill’s vision and decades long experience of creating conceptual works with new electronic mediums and combining this with Süperfad’s digital tech skills.

Reilly had been experimenting with the Kinect platform, and some of the open source performance software that has been developed for it, which lead to their request for custom software to achieve certain ideas. As we worked together developing the software these objectives changed, sometimes due to a limitation we found in the hardware, but also when a new possibility was discovered as we began to understand the toolkit we were working with more clearly.

We continued developing the system for the MARCO performance right up until the morning before the exhibition opening, and spent that afternoon creating video clips (shot and edited by Superfad art director Loren Judah, on the spot). Hours before showtime we frantically entered references to the media files in xml notation, along with gesture definitions, to configure the real-time applications we’d spent the previous month and half creating. Amazingly it worked.

The result was that once the performer was being tracked by the Kinect almost any sudden series of movements created a cacophony of enveloping media. The piece was performed once, recorded as a four channel video, and is now, as of this writing, playing back daily in the Museum gallery where it was recorded. The video recordings were captured directly from the feed going to the projectors in the gallery, providing a time-delayed semblance of the once live performance, sans performer.


The software was written in the Processing programming language using a variety of third-party libraries for that coding environment. Several networked computers ran the software which communicated via OSC to pass control data between them. The MARCO galleries are quite large, this provided a large canvas of white gallery walls to project on. In the end we used five projectors, each fed images by a Mac-Minis networked together via ethernet. The entire hardware system was mounted to a large circular lighting truss that the museum had on hand. For the performance a stage sound system was utilized for audio playback into the gallery.

The first application we built, TextFluidCongeal, was a fluid dynamics particle system where each particle was an individual letter in a phrase. A Kinect, looking down from above and right beside the video projector, tracked movement of feet walking through the fluid. We used the very stable NecTouch by French software artist Benjamin Kuperberg for control data here. NecTouch outputs TUIO, a multi-touch focused specification of OSC, sending the x,y locations of each foot it detects (normally it would be fingers, but we tweaked some settings). We also used Memo Akten’s overly-abused fluid dynamics framework MSAFluid, which, incidentally, receives TUIO quite readily in one of it’s examples for Processing. Some hacking of these two things together, along with several dozen lines of custom code (Processing: the duct-tape of programming languages), allowed us to read phrases we notated in an XML file, and set trigger points that would “congeal” an individual phrase when it’s trigger area was stepped on. A simultaneous phrase would be spoken on the overhead speakers when this occurred by sending an OSC message to another computer on the network to trigger an audio only version of the software described below.

The other application, which I ultimately dubbed MegaFlip, tracked skeleton data and could activate triggers when two or three joints of the users skeleton were in certain range-distance of each other. For example, while the app tracked your movements, touching your head with your right hand could trigger a video clip displayed on a video projector, while touching your right foot with your left hand might trigger an audio clip played back on overhead speakers. Additionally the software could send and receive OSC messages to the other computers so that a gesture tracked by any one computer could trigger play back on any other computer. Each computer’s instance of the application was configured with an XML file. Each node in the XML defined a relationship between joints (i.e. right foot to left hand) and the distance those joints had to be within to trigger a function, also defined in the XML element, along with a reference to which audio or video file to play back and on which computer (by IP address on the local network). Live cameras were also hooked into the software and could be turned on and off via various gestures.

MegaFlip was constructed with multiple Processing libraries as well. SimpleOpenNI was used for interfacing with the Kinect and skeleton tracking. Andres Colubri‘s library GSVideo was used for audio and video playback. Andreas Schlegel‘s oscP5 library was used for OSC communication. Benjamin Kuperberg’s MapiNect was also a great inspiration for our XML mapping schema, as we essentially created a simplified version of this within the software.

During the performance five computers ran MegaFlip simultaneously. The original plan was to have each connected to it’s own Kinect and projector, having them all track the performers skeleton simultaneously and trigger various media. We discovered though that the Kinects interfered with each other when their field of view overlapped too much, and given the circular arrangement of the gallery lighting grid we had to abandon having multiple skeleton tracking computers (though I did manage to get all 5 Kinects track me for half an hour without loosing track during some tests… ). Interestingly the Kinect facing the floor for the text fluid software described above did not interfere with the one Kinect we used for skeleton tracking.

The system was truly put through it’s paces during the actual performance, with dozens of gesture mappings sending signals to play media around the gallery space via the network. The system ran without a hitch during the entire performance, suggesting that it could be stable enough for longer or more complex configurations. It is generic enough to be used for a near infinite variety of unique performances and interactive installations due to it’s ultra configurable nature.