ikarth
commented
9 years ago Well, I'm definitely interested. I don't think that too many people have used prior storytelling AI research code in NaNoGenMo before. I think it's a very good idea.
At the very least it'll be interesting to contrast the original Tale-Spin approach with how contemporary sensibilities lead you to develop your own spin on it. I feel like the increased literacy about cybertext has opened up new avenues for exploration, and I'm curious about what the results are going to be.
PPKFS
rbechtel
MichaelPaulukonis
5Mixer
whitten
I'm in.
At this point, I have no idea what the novel will be about. Instead, I have some ideas about the generation process. I'm going to start with Warren Sack's Common Lisp version of Micro-Talespin (http://eliterature.org/images/microtalespin.txt), mostly for familiarity. I used the Micro-Mumble generation engine from the original Micro-Talespin (converting the original UCI Lisp source to Interlisp) in my dissertation work (Jim Meehan was my advisor), but it's been a very long time, so it's doubtful that any of that past experience will be all that helpful.
I have cheated a bit - I started looking at the code and did a few tweaks as I rebuilt my understanding. I'll include the Sack original and all my "pre-start" changes in an initial check-in after the starting gun sounds. Unfortunately, I don't seem to have my dissertation code, but it would all have to be converted to Common Lisp anyway, so perhaps that's no great loss.
Oversimplifying, I see four development areas: story setup, simulation, story-level decisions, and tactical generation. This is roughly "where do we start", "what happens", "what do we tell the reader and when", and "how do we say it." In Micro-Talespin, there are nearly-separable mechanisms for the simulator and tactical generation, but any story-level decisions are implicitly baked into the code, mostly amounting to "tell them everything in the order it occurs." (The simulator and generator are OK to illustrate the approach, but both will need work to produce something reasonable.)
I'm also interested in looking at using other sources of "what happens", such as event records extracted from news stories (e.g., http://www.gdeltproject.org/) in place of the simulator in Micro-Talespin, but whether the result would be "novelistic" is an open question.
The Talespin approach is OK for an episode. It remains to be seen if it can sustain a novel-length story. The idea of a framing story (as used by others) is one possibility, but there's clearly lots of experimentation ahead.
Yes, far too ambitious. We'll see where things are at the end of November.
Repo is here: https://github.com/rbechtel/NaNoGenMo