The Shape of Strings to Come: A Hands-on Exploration of Grammar-based Approaches to Creative Generation
Linguistic Creativity is the act of squeezing new or surprising meanings from the rules and conventions of pure form. This most democratic mode of creativity demonstrates that the recursive application of declarative symbolic rules – collectively termed a “grammar” – can yield combinations of unexpected novelty and value. While traditional grammars confine themselves to the realm of the syntactic, semantic grammars allow rule-based generative systems to combine anything at all that can be given a symbolic form. Semantic grammars reflect the intuition that many other forms of human ingenuity have a linguistic correlate: consider the written recipes of culinary creativity or the exotic place names of fantasy maps and game worlds. So wherever function follows form, semantic grammars can be used to invent the textual correlates of non-linguistic products and thereby give meaningful shape to those creative products themselves.
This tutorial explores the use of semantic grammars in the creative generation of textual artifacts that range from the primarily linguistic, such as poems and jokes, to the indirectly linguistic, such as recipes and other culinary possibilities. As the technical basis for our explorations we shall use the Tracery formalism of Kate Compton to specify the semantic grammars for our generative CC systems. Tracery is a deceptively simple formalism that permits the creation of elegant generators by researchers at all levels of coding ability, from rank neophytes to experienced developers. Non-coders will find Tracery immediately accessible, while experienced programmers are encouraged to write meta-generators in the language of their choice, to automatically genenerate the grammars that can then generate the end-products of the creative system. The expressive limitations of Tracery (e.g, context-freeness) offer an ideal opportunity to push the generative limits of our tools, and the worked examples of the tutorial will demonstrate how developers can squeeze surprisinging versatility from such simple mechanisms. Participants are encouraged to bring their laptops to work through the examples as they are discussed, to develop their own alternative solutions from the data and knowledge resources that will be provided. As the tutorial focuses on the rapid development of CC systems, participants will unleash their own generative systems on the web as autonomous Twitterbots before the end of the tutorial.
To recap, the tutorial is open to CC researchers and enthusiasts at all levels of coding ability. Bespoke data and knowledge-bases will be provided, and we will work through a series of exemplar generation problems of steadily increasing ambition and complexity. Our systems will range from the merely generative in the first instance to the self-filtering, responsive and personalized in subsequent developments. Teaching materials (datasets, grammars, slides) will be provided from the UCD postgraduate course on computational creativity and from the forthcoming MIT textbook Twitterbots: Making Machines That Make Meaning. Researchers looking to reuse these materials in teaching their own course on CC are also very welcome to the tutorial.
Organizer
Tony Veale, University College Dublin
