Choreographic Language Agent (2009-2013 DOWNIE)

The Choreographic Language Agent project is a small software environment, developed in Field, for exploring variations in choreographic instruction. The promise of formal languages for computers is that, once established, the computer can generate variations of sentences by mutation, selection, and parameterization. In this sense, the Choreography Language Agent can become an active participant in the creative exchange between choreographer and dancer.

The software tool enables the creation of grammars that point two ways — towards simple versions of human language and towards choreographic grammars of dance that are particular to a given choreographer (in the initial case, of Wayne McGregor of Random Dance). This tool posits a new form of dance notation — one which aids the choreographer in generating dance movements rather than in recording existing movements.

Rather than attempting to produce a general model of movement, choreography, and meaning, this tool focuses on the individual and even idiosyncratic methods of a given language movement system. The model takes as its point of departure the minimalist point-line vocabulary of Loops instead of a sophisticated, anatomically correct joint hierarchy, the idea being to rapidly sketch movement explorations at all levels (limb, body, stage-space).

Given a sentence written in a language known to this tool, the agent can interpret this sentence to produce a short animation of its body. Then we can perform pseudo-linguistic operations on the language level, thus generating sequences, superpositions, and modulations. After studying sets of sentences, the choreographer can determine the sets of conditions under which an agent can autonomously deploy this language to fashion a multi-agent piece of choreography.

The results are interesting to us for three reasons:

  • the way in which the tool provokes and perturbs choreographic decisions is informed by the practices of representation and manipulation of a number of choreographers; besides McGregor, we’ve seen related approaches in Forsythe, Cunningham, Jones, and Brown. Each of these choreographers create cold formal systems, the operations of which provoke new ideas for movement. Their actual choreography arises from human dancers’ interaction with this formal system.

  • the project offers an alternative to existing digital tools, which operate at too concrete and literal a level: if you show dancers a traditional keyframed body animation, it is hard for them not to simply imitate its movement directly rather than to invent strategies for reinterpreting the movement it shows onto their own bodies. the project could lead to a fascinating virtual dance that is intricate at a number of levels, from the details of individual movements to the over-all picture of their synchrony.

  • the automatic translation of language into different manifestations (here, movement) and back again is core to our central interests (c.f. Breath and Other Bodies).


The core of the idea is that you can start creating choreography by writing sentences in a formal language (in this case, inspired by a reading of words that Wayne MacGregor has used to choreograph a dance). The agent then automatically translates that sentence into animation — specifically, it animates the points constituting the dancer’s body, the points corresponding to the room space, and the points of the dancer’s kinesphere (actually a cube).

New words can enter the language in three ways: first, you can simply program them in using Java or Python; second, and more interestingly, you can refer to collections of points that you have selected and named; and third, and most interestingly of all, you can refer to diagrams that you draw directly on the screen in 3D.

With language at its core, the agent can use language to incorporate other materials into its world. We imagine that in addition to incorporating diagrams and point collections, the language could also accommodate other sources of motion — most promisingly, motion-capture data sets.

Choreographic practice proceeds from small to large. At the smallest level, you have individual sentences, which you can scrub through and then alter by editing the text. Multiple overlapping sentences can then be combined on a single timeline to construct a visual score in 3D.

Soon the agent will support two further levels. The grammar will include the specification of conditions in which sentences may be activated — thus forming a truly autonomous agent that makes decisions for itself according to the rules that you write.

In addition, the tool will soon support choreographic sentences for multiple figures.


Wayne McGregor and his dancers have been working with the Choreographic Language Agent both in development sessions and in rehearsal, most recently to create Undance in 2011.


This project has been supported by Portland Green Cultural Projects.