Tuesday, October 30, 2007
Another step on the stairway of surprise
More tests:
Returning jealousy
The brittle ladies that keep
and unroll, and a fiery
sin, an invincible sin
A brittle lady that understands and realizes,
and the unrestful
madams
The brittle ladies that
behold and unroll, and an
ultimate gentlewoman
A brittle lady that understands and emerges,
and a broad sin, a
marauding sin
Tuesday, October 23, 2007
Just when we thought we were making progress
Etc3 conceptualizes the poem as the end product of three contributing models: a semantic source model, a grammar, and a surface realizer. This has worked pretty well as the system expands.
But....
So far my emphasis has been on the lyric, with Sylvia Plath's poetry as grammatical model (and again, I don't care for her work--but it does evince recognizable markers that make for easy translation into a grammatical description). There is in this
description a clear separation between the system's grammar and the final formal realization of the poem. There are several levels of realization implemented in the system, some at the node level, some at the tree, and then the final presentation
layer, the stanza. As it turns out, you can actually untether the lyric's formal presentation on the page from the grammar of its utterances (individual tag trees). Different arrangements of the lines on the page seem not to distort the "poeticalness" of the piece.
So I've implemented a class hierarchy for the stanza, with an abstract base class with subclasses for different presentations. A side benefit of this approach is that the stanza built in this way can be thought of as the root node in a tree from which
all of its utterances branch, and different grammatical structures can be flagged as appropriate for different positions within the poem (open, close, etc.), which, at first seems like a grammatical description but is really compositional arrangement. This approach has worked well enough up to now.
But...
I've turned to building another grammar (Etc allows for multiple grammars), this time modeled on Rachael Blau DuPlessis's Drafts. And the conceptual structure that worked for the lyric won't with her work.
Here is a recent Etc lyric (composed during a test of some feature or other--I don't recall what):
Like a proof
There is no mathematics more
lost than love
I do not see her
water, her peace, her rest
It is I
who become her
My throat a sea
in the depths and penurious enough
to guess
The vast wrists, little
wrists, huge wrists of a broken
sail, like conscious
toils
The dead prizes offer
Someone toddles a craft, where sails and
eyes and transports
bring sombreness
Where mathematics brings its love
"Like a proof" is just as effective in a different form:
Like a proof
There is no mathematics
more lost than love
I do not see her water,
her peace, her rest
It is I who become her
My throat a sea
in the depths and
penurious enough to guess
The vast wrists, little wrists, huge
wrists of a broken sail, like conscious
toils
The dead prizes offer
Someone toddles a craft,
where sails and
eyes and transports
bring sombreness
Where mathematics brings its love
So far as generation is concerned, "surface" in poems like this is more convention
than substance. But not so with this fragment from DuPlessis's Draft X: Letters:
and me at my hieroglyphic Or for orphic ears, to hear,
Mac, as high or orphic, orphaned mouths,
toned as Xerox, "oh oh"
A well with clefts, letters as stones.
Formal presentation here matters and is inextricably welded to the text's underlying grammar, so much so that surface structural considerations are equal manifestations of that grammar, just as much as are the lines it presents. Where intuition says that radical poetry like this ought to be easier to generate via a controlled randomization of its recognizable markers, it's not. Not at all. It's much more difficult.
And it has enormous ramifications for extending Etc to include emulations of such work. Whereas up until now, adding new base classes to the stanza hierarchy was straight forward and really quite simple, it won't work here. Emulating DuPlessis will require an equally radical rethinking of the current design.
All complex (even moderately complex) systems seem to hit this barrier. We (at least I) never seem to be able to anticipate all of the uses to which we want to apply the emerging system. And that carefully constructed apparatus for extension and enhancement of which we are so very proud threatens to collapse, reminding us yet again that we only think we know how to program. In the end system design and implementation is not only a humbling, but a humiliating profession.
