In an earlier post I mentioned that each video episode in my proposed project will have three keywords, and that one of these will be a “Bachelardian” label. Sounds all mysterious really. Gaston Bachelard is/was a French phenomenologist who wrote some works that are seminal for me. The Poetics of Reverie, The Poetics of Space, The Psychoanalysis of Fire, The Poetics of Matter (you get the drift?). It is a long list.
So my labels, I don’t actually know what the words will be yet (the actual words are irrelevant, they’re just values in the system), but the paired terms are:
- stasis – action (movement)
- soft – hard (matter)
- fire – ice (hearth)
- high – low (space)
- near – far (touch)
After compiling that list (which I just brainstormed) I realised that these are the only keywords I need in the project. Any project will probably have at least two of these, and if I build the project in such a way that it will first of all search for a match on any of the keywords, and then add a random connection half way through each episode (if it doesn’t find sufficient matches) then the structure should be fine.
However, that probably needs a bit of backstory, doesn’t it? The Korsakow Engine uses keywords. These are defined by the author and have the usual complement of Boolean conditions. Hence I can attach keywords to an episode (and impressively can have keywords attached to points in the timeline) and then for the current episode search for other episodes that on the basis of whatever keywords I wish. Hence the current episode might have the keyword ‘yes’, and I can get it to search for episodes that have the keyword ‘yes’ and ‘no’, for example. I can exclude terms, and so on. The way I am using the engine there is one episode which plays, and then three selections are made (on the basis of the keywords) and these appear as thumbnails across the bottom of the screen. Clicking any of these makes this the episode that plays, and in turn it uses the assigned keyword values to determine another three episodes to appear below.
Now a very common error in authoring systems that have such an architecture is to over script such a system. Assigning lots of keywords, lots of rules. This generally makes it hard to author (as you have so much to map or manage), or it makes the structure just irrelevant to the reader. After all, it is the reader who is crucial here, and if the reader can’t discern a pattern then the work will not be understood, and (as is usually the case) dismissed.
Hence in the project that I’m slowly piecing together I’ll just use the single set of keywords, assigning as many as are relevant per episode. Repetition is fine, and helps create the patterns that allow structure to emerge (repetition after all is fundamental to pattern). If there aren’t any relevant keywords, and to add a bit of noise into the system (to avoid building constrained loops that you can’t get out of), each episode when it makes a selection of related episodes can randomly select a third (if required). This happens half way through the current episode so won’t appear as a possible selection until some time interval.
I could make the structure much more, well, structured, but as I want to be able to continually and easily add new content I don’t have the advantage of a known body of episodes. So a very simple keyword system, combined with one random element, should generate enough connections to provide a navigable and intelligible structure. This is basically a constrained taxonomy, and while each episode may be highly self contained I think that once there is sufficient content in there connections will appear between the segments courtesy of the relations the system generates and the fact that all the work is mine. We’ll see.