We Can Re-Use It For You Wholesale: Serendipity and Objets Trouvs in Linguistic Creativity

The objet trouvé or found object has become a staple of modern art, one which demonstrates that artistic creativity is just as likely to arise from serendipitous encounters in the real world as it is from purposeful exploration in the studio. These readymades , like Duchamp’s Fountain (a urinal!) seem banal in their conventional contexts of use, but take on a new resonance and meaning when viewed in an artistic space. This paper considers the linguistic equivalent of the readymade object: a well-formed phrase that takes on a new significance when used in a new context with a potentially new and resonant figurative meaning. We show how linguistic readymades can be recognized and harvested on a large scale, and used to provide a robust and scalable form of creative language generation. For Best Results, Just Add Water Computationalists in the classical AI tradition generally prefer to model the creative process as a purposeful exploration of a conceptual space (e.g., see Boden, 1994). This concentration of computational effort makes good engineering sense, but little artistic sense, since much of what we consider to be creative insight occurs not in the studio or the laboratory but through everyday interaction with the real world. Serendipitous discovery is thus unlikely to arise in purposeful explorations, since specific applications have no remit beyond the immediate concerns of their programming (and programmers), and have no other lives to live when not actively pursuing these concerns. The objet trouvé or found object is perhaps the most potent example of serendipity in artistic creation. An artist encounters an object with aesthetic merits that are overlooked in its banal, everyday contexts of use, yet when this object is moved to an explicitly artistic context, such as an art gallery, viewers are better able to appreciate these merits. The transformational power of a simple context switch is most famously demonstrated by the case of Marcel Duchamp’s Fountain, a humble urinal that becomes an elegantly curved piece of sculpture when viewed with the right mindset. Duchamp referred to his objets trouvés as “readymades”, since they allow us to remake the act of artistic creation as one of pure insight and inspired recognition rather than one of manual craftsmanship (see Taylor, 2009). In computational terms, the Duchampian notion of a readymade allows artistic creativity to be modeled not as a construction problem but as a decision problem. A computational Duchamp need not explore an abstract conceptual space of potential ideas. Rather, the issue instead becomes: how do we expose our Duchampian agent to the multitude of potentially inspiring real-world stimuli that a human artist encounters everyday? Readymades represent a form of creativity that is poorly served by exploratory models of creativity, such as that of Boden (1994), and better served by the investment models such as the buy-low-sell-high theory of Sternberg and Lubart (1995). In this view, creators and artists find unexpected or untapped value in unfashionable objects or ideas that already exist, and quickly move their gaze elsewhere once the public at large come to recognize this value. Duchampian creators invest in everyday objects, just as Duchamp found artistic merit in bottles and combs. From a linguistic perspective, these everyday objects are commonplace words and phrases which, when wrenched from their conventional contexts of use, are free to take on enhanced meanings and provide additional returns to the investor. The realm in which a maker of linguistic readymades operates is not the real world, and not an abstract conceptual space, but the realm of texts: large corpora become rich hunting grounds for investors in linguistic objets trouvés. This proposal is realized in a computational form in the following sections. A rich vocabulary of cultural stereotypes is acquired from the web, and it is this vocabulary that facilitates the implementation of a decision procedure for recognizing potential readymades in large corpora – in this case, the Google database of web ngrams (Brants and Franz, 2006). This decision procedure provides the basis for a robust web application called The Jigsaw Bard, and the cognitive insights that underpin The Bard’s conception of linguistic readymades are then put to the empirical test using statistical analysis. While readymades remain a contentious notion in the public’s appreciation of artistic creativity – despite Duchamp’s Fountain being considered one of the most influential artworks of the 20 century – we Proceedings of the Second International Conference on Computational Creativity 22 shall show that the notion of linguistic readymade has significant practical merit in the realms of text generation and computational creativity.