Can computers write stories?
Can computers write stories?
Dateline: 02/12/00
Selmer Bringsjord, associate professor of philosophy, psychology, and cognitive science at Rensselaer Polytechnic Institute, and David Ferrucci, senior scientist at IBM's T.J. Watson Research Center started working on the problem of building a sophisticated artificial storyteller in 1991. The end result of this $300.000 worth effort is Brutus.1, "the world's most advanced story generator" that can autonomously write short fiction of up to 500 words around the concept of betrayal. It has knowledge about the ontology of academia: professors, dissertations, students, campuses, and needs a formal mathematical definition of betrayal. However, any question outside its knowledge base would draw a blank as it requires pre-programming of the whole system. Anyone interested in describing the concepts of love, revenge, jealousy to the computer so it can write truly compelling stories?
Researchers expect that in the future the entertainment industry will rely on such artificially intelligent systems that know how to create and direct stories. Seems that the answer to the question from the title is positive - nonetheless, Bringsjord admits that computers will never best human storytellers in even a short story competition. He will continue to build descendants of Brutus.1, but does not expect a machine that could understand the "inner lives" of his or her characters. To do that, it would need not only to think mechanically in the sense of swift calculation, it would also need to think experientially in the sense of having subjective awareness. Could we build such machine? Believers in "Weak" AI (like Bringsjord and his colleagues) say that human thought can only be simulated in a computational device, and not broken down into a series of mathematical operations and algorithms (as expected in the opposed, "Strong" branch of AI). Taking this idea a step further, we may have machines that will exhibit much of the human behavior, but none of these machines will ever be more human than an average vacuum cleaner. These philosophical issues recently sparked a great discussion in our Forum, so if you are interested please share your thoughts with other readers.
Returning to the more practical level, the related subject of Natural Language Generation (NLG) is receiving more attention these days. Until quite recently, most work in Computational Linguistics concentrated on the interpretation task. To put it formally, and define by the words of D. D. McDonald, NLG is the process of deliberately constructing a natural language text in order to meet specified communicative goals. Using automatic text generation has a number of advantages allowing us to automatically author texts on demand, tailored to the context of use and the user's needs, and taking into account the most up to date information available. As with almost any other AI-related topic, there are several excellent and up-to-date information sources on the Web that can provide you with more information on NLG. Let's start with a repository for information about Natural Language Generation (NLG) research in Scotland. I don't know the reason for a particularly high concentration of NLG people in Scotland, but the pages are really informative. German researchers built a (very) similar site dedicated to NLG resources in their country. At the other end of the world, MRI's Language Technology Group is offering a number of simple prototype natural language generation systems running on the Web which serve as testbeds for exploring their ideas: Power, Peba-II and StockReporter projects. Similar set of project reports can be found at University of Brighton describing DRAFTER, GIST, AGILE, RAGS and GNOME. The Natural Language Software Registry will provide you with a set of links to the related software packages. Erasmatron represents an entirely different view in interactive storytelling.
Association for Computational Linguistics Special Interest Group on Generation is a frequent stop, as it provides a forum for the discussion, dissemination and archiving of research topics and results in the field of text generation. Who's who in NLG and What's where in NLG will save you hours of searching, as they offer condensed and easy to use presentation format.
Another good one-stop resource is a book titled "Building Natural Language Generation Systems" written by Ehud Reiter and Robert Dale. It may be of interest to students and researchers interested in NLG, and also to software developers working on advanced document-generation systems. Another book, perhaps more suitable for the layman is Roger C. Schank's "Tell me a Story: A new look at Real and Artificial Memory". It offers insights into the nature of stories and their importance to intelligence, memory, understanding, and knowledge.