I’ve always wanted to make some interactive fiction, and last month I decided to spend some time working on a procedurally generated Inform7 game. You can read more about it on the Github “Issue” thread where I announced my intent to participate, but I’m reposting my final “wrap-up” entry here:
Post-mortem / wrap-up post:
Failures
Mainly, I didn’t make a 50k word “story”. Although you could maybe argue that the inform “story” isn’t the “whole story”, that’s what my script generates, and I was planning on using that for the word-count metric. My plan for getting to 50k had basically been to increase the number of rooms the script generates until I hit that number, but that isn’t possible because some of the items I’m using for randomization don’t contain that many items. (And I didn’t realize or notice this until just now as I was prepping this post!) Turns out I can only push the script to ~25 rooms, before it tries to randomize from several sources with only that many items. The word count at 25 rooms is only about 4700-4800.
Also worth noting that version with 25 items doesn’t always work. I went ahead and pushed an example of this output as `output/v0.5-source-broken.txt`. Pasting that script into inform gives errors that I’m probably not going to take the time to fix. (Essentially, some of the source for my randomized room text is probably problematic, and should be excluded.)
Additionally, I’m pretty sure the game as-is at the moment of this writing isn’t that fun to play. There isn’t enough randomization of the puzzles. Essentially, each room has the same fetch quest. (If you’d never played it before, and went into it without my spoiling it for you –which I’m not going to do here– it’s possible it would take you a bit to figure out, but once you did, you’d know how to solve every other room, and it would grow tedious pretty quick. I think the version with 4 rooms (in a 2×2 grid) is probably the best way to play. There is some novelty in the randomized room names and descriptions which can sometimes be pretty surprising. It might be fun up to 12 or 16, but again, it would get old pretty fast in its current incarnation.
Another failure, I could argue, was my goal of learning Inform7, but I’ll write more about that in a bit.
Because I wasn’t really happy enough with the output of this script to post it anywhere, I also didn’t submit it to ProcJam. That was/is also definitely a fail.
Successes
I wrote some Python! Python is ridiculously easy to write. It feels a bit like I say this about every new language I learn these days, but learning the syntax (there is almost none!) and the various APIs was a very minor part of this project, and generally quite fun and easy. Debugging errors was quite easy, as error messages were very easy to search for, and often even that step wasn’t needed.
I learned quite a bit about writing Inform7! Unfortunately, that’s about the best I can say about it. Inform got harder and harder to work with, and was generally the opposite of my experience with python. My take-away from Inform is that you want to write one sentence at a time, compiling after every one. Every new thing you try and do will require searching through the documentation for an example you can copy/paste. There were dozens of times when I would modify even just one word from an example and then scratch my head about why that changed caused it to stop working. And debugging was always a nightmare. The error messages sometimes contain (sometimes hilariously) a bit of randomization themselves. This seems interesting/funny the first few times you see the same error and the text is different, but the 3rd or 4th time, when you are at wit’s end, it’s no longer funny. Here’s an example of an inform error (but not one with randomized text, I don’t think):
You wrote ‘now the Greyish Blue Book Of Matches is nowhere’ , but although ‘the Greyish Blue Book Of Matches is nowhere’ is a condition which it is legal to test with ‘if’, ‘when’, and so forth, it is not something I can arrange to happen on request. Whether it is true or not depends on current circumstances: so to make it true, you will need to adjust those circumstances.
In general, my take-away is that Inform7 syntax is a great big can of worms. It would probably take me a month of working full-time to really understand the entire system and how it all works together, and feeling competent in it would probably take much longer. (This was obviously not that month!)
Wishlist / TODO list
If I wanted to spend more time on this project, these are the things I was planning on doing:
* Randomized puzzles. Right now there is really only one type of puzzle. I would love to have 4 or 5 types (at least!), and generate different room descriptions based on the type.
* Additional inform elements: Right now, there are rooms, objects (some edible), containers, and that’s about it. I really wanted to get to the point where there was also a randomized person in each room. The amount of things you can do with Inform is staggering really. I’ve only just scratched the surface, for sure.
* Finally, making this available to the rest of the world. This boils down to how I wanted to do this initially, and how I think it’s feasible. Both would have been published as a webpage. A) I wanted to make a version that was different every time you play it. But the only way I could imagine that happening would be to install the command-line version of inform on a server, and at request time, generate the source, compile it, and redirect the request to that output. I’ve no idea if that’s practical, but it’s not something I wanted to attempt. B) The more practical alternative would be to generate a bunch of outputs all at once. So that would be writing another script to run my `game.py` script X number of times, saving the output off to a tmp directory, then using the command-line version of Inform7 (i7) to compile each output and save off a web-playable version into a subdirectory somewhere. Tentatively I was planning on doing this 365 times and then writing some kind of index.php to swap them out based on day of the year.
TL;DR
I really enjoyed working on this project, and learned a lot about Inform7 and python, but wouldn’t really consider this a “successful” project, mostly because I just didn’t spend the required time on it. There is always more you can do, of course, but in this case, I didn’t take the project far enough where I think it’s ready for public consumption. (It is all public, however, and anyone can play with the stuff I created. If you do, I’d love to hear about it!)
There might be better/additional corpora resources with NLTK http://www.nltk.org/
I believe TL;DRs go at the beginning of an article, you basically tricked me into reading this whole thing.