The AAAI workshops and symposia that I’ve been to, which are more focused on the topics we’re discussing (e.g., Intelligent Narrative Technologies) have actually had a good number of interactive systems in play.

I’ve been to the main ACL conference and gave an invited talk at NAACL two years ago. There is a definite preference for non-interactive systems there, yes.

It seems to me that these super-small systems
do something poetic without being interactive.

Nah, I wasn’t hating, about half my generators are non-interactive. I was just thinking if I were to write such a program myself I’d try to save characters by prompting the user for data rather than storing it in the program.

there is something of a myth that an interesting
system has to be interactive

Maybe in the digital humanities and e-poetry communities, I’m not familiar with those. But if you look at the proceedings of ACL or AAAI, the overwhelming amount of research seems to occur in non-interactive systems. And some of the most interesting research (in my opinion) is being done with interactive systems in relatively non-mainstream communities like SIGDial (dialogue systems) or HRI (human-robot interaction). Lots of inspiration there for poetry generators, both interactive and non.

thx for the papers!

All of the ppg programs are non-interactive; they aren’t machine learning systems so neither supervised nor unsupervised learning is involved. I don’t have any distaste for interactive systems, but there is something of a myth that an interesting system has to be interactive. It seems to me that these super-small systems do something poetic without being interactive.

The ppg256 programs do include data and, as I discussed some at Stanford, parameters as well. By saying that they aren’t “data-driven,” I meant to indicate that they aren’t like machine learning systems that work by being directly trained on data, by being pointed at corpora and documents. Many of the ppg256 programs and “Sea and Spar Between,” which I wrote with Stephanie Strickland, did involve computer and human analysis of existing corpora, but the source texts in these cases were structured, curated, etc. in many non-automatic ways.

I don’t have any paper publication covering the same topic as my Stanford talk, but I’ve written a bit throughout the years – and am writing more currently – on this question of the humanistic study of code. For example:

“A Box, Darkly: Obfuscation, Weird Languages, and Code Aesthetics” by Michael Mateas and Nick Montfort.

And I have a paper on the ppg series:

“The ppg256 Series of Minimal Poetry Generators,” Nick Montfort.

I was very pleased, on June 22, to join a panel…

Interesting points. Just out of curiosity, is interactivity compatible with your ppg approach, or are you committed to unsupervised generation?

ppg256 and Concrete Perl, which are not data-driven

If I may nitpick… When I look at (for example) ppg256-1, things like “cococacamamadebapabohamolaburatamihopodito” sure look like data, and they clearly play an important part of the words that emerge. I think you mean something like “hand-authoring linguistic resources rather than exploiting existing corpora”.

as humanists we should be “digging into code”
as well as data, understanding process in the
new ways that we can.

I’m interested in what you mean by this. Have you written it up somewhere?