Search This Blog

Follow adrianbowyer on Twitter

My home page

Thursday 24 February 2011

NeuroBuster


It is notoriously the case that nobody in Hollywood has a clue what they are doing.

Specifically, nobody can tell the difference between a script that will be a blockbuster and a script that will be a turkey.  Possibly I should moderate that: maybe some scripts are definite turkeys.  One can't imagine that a script written by a six-year-old would make a fortune.  Or maybe it would...

Anyway.  From among the remainder, nobody can tell.

Perhaps it's time to take people out of the choice altogether.   I have written here before about stylometry: the battery of computational techniques used to compare the authorship of documents.  These techniques come up with a large number of statistics on a given document.  Comparing those statistics with the values from another document allows a probability to be assigned to the hypothesis that both were written by the same person.

But that's not all that one could do with such a large bunch of stats obtained from - in this case - a film script.  They could also be used as input to an artificial neural network.

Artificial neural networks are computer programs that learn to distinguish between sets of input parameters.  For example, you might want a computer program to examine CCTV footage from a road and automatically to flag up when there was a traffic jam on it.  In this case what you would do is to show the network lots of historical footage of the road when it was clear, and another lot of when it was jammed.  As the two sets of historical teaching footage have already been classified, the network can be taught to distinguish the two and to flag up one of two outputs: clear or jammed.

The stylometric measures from movie scripts are just as good a potential input to a neural network.  Hollywood has hundreds of thousands of old scripts, and of course it knows the blockbuster or turkey status of each.  You use those as the teaching input.  Once again the network has two outputs: produce this movie, or bin it.

Large numbers of expensive movie executives and their acolytes could be dismissed, and the whole problem of movie finance could be turned over to a PC.  When it lit up and rang its bell, a nearby secretary would pick up a phone and call a director from a list.  Job done.

Of course, this doesn't end with movies.  It would work just as well with TV scripts, and with novels.  Publishers would no longer have to flatter the pampered egos of their once-bestselling authors in the futile hope that they will be able to do the magic again.  All manuscripts, whether from the famous or the unknown, would go into the machine, and out would come a whittled-down but steady stream of copper-bottomed, sure-fire publishing sensations.

No comments:

Post a Comment