Summing it all up.

I don’t want to spend too much time running over our Fever Dreams. We’ve already discussed some of the positive and negative experiences that came up during the Fever Dream process, so if you’re reading this after our previous entries you’ll know all about them. And if not, maybe it’ll give you an excuse to look through those older posts.

In this last one, I wanted to share two images that demonstrate what I’ve alluded to throughout the project – the odd way our automated summary tool dealt with scripts of varying length & quality. The images feature the same data but presented in slightly different ways. (Hopefully that will stimulate some additional thought). I follow those visualisations with a quick discussion about one “lesson” we learned from this whole shindig.

Visualising our outputs.

The first image (see below) shows the change in order when examining a film’s IMDb rank, its script word count, and the resulting word count of the Fever Dream script. As discussed before, the automated summary tool appeared to react in an unusual, non-replicated way each time it faced a new script. That oddity is demonstrated best if you follow the paths of Shawshank (the only film with no order change between the original and Dream word counts), and Godfather & Fight Club (the two films with the greatest changes in order, downwards & upwards respectively). It is a simple depiction, but it gives the distinct feeling that some wires may have been crossed.

Our second image is a planetary exhibition that shows the same information as the alluvial diagram but in a more active manner. When you watch the gif below, you’ll find each film’s IMDb rank represented by its place in the solar system – the closest to the center, the higher its ranking. The size of each planet corresponds to the word count in the original script (the larger the planet, the longer the script), and the planet’s orbiting speed relates to the word count in the Fever Dream script. The longer a planet takes to revolve, the longer its Dream script. The shorter its Dream script, the quicker it moves around the center.

As seen above, (and as discussed in our previous entry) the slowest planet is also the tenth ranked film, i.e. Fight Club. The gif is set to follow one revolution of Fight Club before resetting (this lasts about 45 seconds), and it’s somewhat interesting to track the number of revolutions the other planets manage to fit in. (The Godfather gets nearly three).

Learning a lesson or two (but really just one).

It would be superfluous – to the point of absurd – to do a “post-mortem” on our Fever Dreams work. So rather than listing out all of our woes, I’ll keep it brief and limit this last bit of discussion to the one change with the greatest potential for improving our approach.

Who cares about the film rating?

When I started Fever Dreams, I thought it best to simply work my way down the IMDb’s list of top-rated films and produce a Dream for each one. By doing so, we could examine a collection of films that spanned different genres, styles, and principles of storytelling. In theory, this would provide the summary tool with a unique challenge each time; summarising a political thriller might not work in the same way as a superhero film. (This does rely on the assumption that a list of top-rated films actually represents multiple film types). But whether or not the films were “representative” should have been a secondary concern to the choice of scripts. After all, the scripts are the material we feed into the summary tool.

Of the ten Fever Dreams we put together, four had missing scenes (i.e. scenes in the script that were not present in the film). While this was not an insurmountable issue – I still managed to cut together the Fever Dream versions of the film and discuss the plotlines presented – it did cut against the basic idea behind the project. I may be talking in circles, but we wanted to summarise a film based on the summary of its script, and it’s a real issue if the script summary has details that are not contained in the actual work (i.e. the film) we want to summarise.

In the grand scheme of things – as grand as this is – it would not have mattered if the films were ranked 2, 18, 55, 68, etc., as long as the scripts for each were formatted the same way and were, as best as possible, the actual screenplays for the films.  Although it would have taken additional time, some of the limitations to our approach could have been addressed if we had selected a base set of scripts (all from the same source) and then watched the films while comparing them to those scripts. We could have then removed the extraneous scenes from the scripts or used only those that perfectly mirrored the film. Either way, we would have been left with an internally consistent group of scripts and could then have selected various kinds of film for the summary tool to confront.

And that, I think, is the biggest lesson for anyone to consider (probably just me) if they were going to do something similar. In retrospect, it seems fairly obvious!

Thank you!

If you made it this far – in either our entire Fever Dream journey or just this post – I’d like to say thanks. Who knows if anyone enjoyed them, but I’m hoping that someone did. And if not, life goes on.