Page 90 - AC/E Digital Culture Annual Report

90
cases” – allowing the application of solutions that have worked in the past to solve current problems – to construct new stories based on the reworking of structures, components or fragments of existing stories. The rest of this section will take a look at a series of research projects related to the creation of literature from a computational point of view that includes endeavors in the generation of both automated poetry and narrative.
The invention of a poem by computer involves reworking large fragments of pre- formatted text or generating sequences of text through a procedure that guarantees the continuity of the words.
Generating poetry
The invention of a poem by computer has tradi- tionally employed two strategies. One involves reworking large fragments of pre-formatted text in lines structured like a poem. The other is to generate a sequence of text through a procedure that guarantees the continuity of the words and then establish a distribution of the resulting text in lines using an additional procedure of division.
The reuse of fragments of text already set out
in a poetic structure was started by OuLiPo (Ouvroir de Littérature Potentielle or Potential Literature Workshop), an experimental literary group set up in 1960 by predominantly French writers and mathematicians who were seeking to create works using specific writing techniques. This literary trend considers potential literature as the search for new forms and structures that could be used by writers as they see fit. To this end, the use of computers allows for the explo- ration of different ways of combining existing fragments of poetry, such as taking the structure of Baudelaire’s poems and replacing the words with others taken from the poems of Rimbaud.
More recently, techniques of this kind have been used to generate poems as part of the investiga-
tion into computational creativity. Although ini- tially this was done mainly in English, the range of languages has gradually broadened and now poetry can be generated in Spanish, Portuguese, Basque, French, Indonesian, Bengali, Chinese, Mandarin and Finnish. The investigation under- way explores the means in which techniques, such as the construction of workflows, restricted programs, analysis of the body of work and the representation of analogies, can help to improve the results already achieved using this method. In some cases, the technique has been refined to the point that the examples taken from original poems are completely emptied of words, leaving nothing but the corresponding syntactic catego- ries to be filled with equivalent words taken from other sources.
Other procedures are based on constructing
a sequence of text from zero, using specific techniques to guarantee that the resulting text will be linguistically correct or at least have a certain semblance of fluency. By generating text like this, the results are always grammatically correct. A more popular alternative is the use of n-grams to model the probability that certain words will follow others. Using a body of various texts as a reference, every time the question arises of which word follows the given word, one is selected based on the body of texts. This guarantees that the resulting word sequences are never incongruous while tending to generate texts that lack meaning.
Generating narrative
The automatic generation of narrative has been tackled from two different points of view: as a process of generating plotlines for new stories (inventing new stories) and as a process of relating how the events of a story have occurred (storytelling). The automatic construction of the plotline of the story is based on two different sources: the AI techniques and existing narrative studies.
COMPUTER-DRIVEN CREATIVITY STANDS AT THE FOREFRONT OF ARTIFICIAL INTELLIGENCE... · PABLO GERVÁS
Digital Trends in Culture