On Motivation and Expected Impact

The StoryTelling Laboratory brings together a group of researchers interested in exploring ways of getting machines to handle stories and language the way people do. We have been working on this now for more than 20 years.

We have discovered that, because people use stories and language in many different ways and for many different purposes, such a simple goal becomes a considerable undertaking that involves many different tasks and spans several disciplines.

We have discovered that there is no single discipline that holds the right tools or theories to address these tasks. We will consider any tool from computer science or artificial intelligence if it addresses an issue we are working on, but we will not be lured into considering that it holds the golden answer to all the problems we are facing. We will consider any theory from narratology or psychology if it informs an issue we are working on, but we will not be shy of departing from it if it at some point useful solutions to the task require it. We will not be swayed by academic trends or black lists. We will not be afraid of combining technologies or theories whenever that pushes us closer to our goals. We have found that, while this clearly hinders progress in specific academic communities and leads to low results in research assessment rankings, it enhances the chances of meeting our goal. So we are happy to pay the price.

The nature of academic research is ill-suited to long term commitment to specific goals. Funding and manpower restrict what can be done, and both come and go over time. You will find that the list of topics covered below is broad, and yet many have only been addressed at a very preliminary level. This is because: (1) the field of topics that would need to be covered to satisfy our goal fully is immense, (2) we start on a new topic whenever we find someone interested on working on it, (3) people are often driven on by their academic careers to new horizons, leaving their work on these topics at an early stage.
We walk a tightrope between an engineer’s wish for the simplest posible solution to the simplest version of the problem and a humanist’s desire to find answers to the most complex refinements of narrative usage or psychological nuance in literature. We are proud when the basic prototypes for a clearly defined subtasks perform well, and yet we suffer because the subtask is only a small part of the problem and the results show the lack of everything we have put aside to get there. So we try to address different subtasks, in the hope that an overall view of the field will progressively emerge.

The work reported here does not intend to be a grand theory of computational narrative or a marketing place for new AI technologies. It is simply a log of work done in good faith on a very difficult challenge with very limited resources. It is very obviously a work in progess and we suspect it will remain so for a long time to come. As we move forward we will try to make sense of the accumulated discoveries, but a snapshot at any point in time is likely to have gaps and inconsistencies. As in any log not subject to military discipline, there will be periods when we are too busy working to update the log regularly. This has indeed been the case for long periods in the past, and it will very likely happen again. If you find that this log has not shown progress for some time, please check the list of Publications in the NIL research group web site (http://nil.fdi.ucm.es). That tends to grow steadily even when we do not have time for elaborating on our work here.