I ♥ E-Poetry

This is part 5 of a series of 5 postings on poems generated by the “Frequency” program. For an overview of the series, read the first entry.

The exercise of creating a data set based on Oulipian constraints and creating array and mechanisms to generate certain pre-existent forms, laid the groundwork for Rettberg to be inspired to create three forms: Doubling (discussed in the previous entry), Four Square, and Two Towers. The two latter forms were powered by three of the Machine’s capabilities: the ability to count characters in a line (including spaces), use that information to generate arrays and poems, and use a fixed-width font (Courier). This gives Rettberg pictorial control over text, allowing him to create relations between lines of the same string length and cluster them accordingly to create stanzas.

The Four Square poem depicted above has four four-line stanzas selected by character length, rather than by the more traditional methods, such as metrical feet, rhyme, or syllable count. Because they are generated from an automated array it doesn’t really show how many lines fit a particular character count. This may account for some of the line repetition in the first two stanzas, which lends the poem a refrain-like musicality. Their indentation, another variable, provide added interest in the stanzas’ spatial relations. Are the two central stanzas embedded within an idea expressed in the first and last stanzas? And why is the last one more indented than the first?

The moment we start asking why, we come to the question of intent— why Rettberg made certain choices. If you seek a satisfactory answer you have to resist the urge to state that these particular indentations and choices are meaningless because they, like much about this poem, are randomly generated. These poems aren’t merely Turing tests, attempting to fool readers into projecting a false authorship onto them (though I wouldn’t put it past Rettberg). A more productive approach is to think why Rettberg made the choices he made, including the choice to leave some decisions in the hands of his collaborator, the machine. And his intentions are encoded into those processes. As for the stanza indentation is concerned, I think he placed it there to suggest alternate relations between stanzas than the single train of thought one usually gets when reading a traditionally formatted sequence of stanzas.

The Two Towers poems generate two character-based stanzas next to each other that add up to 40 characters, as can be seen in this random example:

This form presents multiple ways of reading these stanzas: individually, sequentially, or line by line (across the divide) to form new line combinations, call and response situations, and the juxtaposed voices of two different kinds of lines.

It is probably something he might not have come up with, had he not embarked upon the “Frequency” project. Here we witness digital poiesis— a true collaboration between Rettberg and the Machine.

This is part 4 of a series of 5 postings on poems generated by the “Frequency” program. For an overview of the series, read the first entry.

This entire project was inspired by the OULIPO, and how they created new poetic and literary forms based on mathematical and other constraints. Not only did they create a great number of fascinating constraints— such as the n+7, the snowball, and all manner of permutational works— they also helped us rethink traditional poetic forms as constraints that led to creativity. From this perspective, a sonnet or a haiku are little rhetorical machines, and to write one is to shape’s one’s thoughts and language choices to capitalize upon their affordances and constraints.

The poems focused on in this entry are inspired by the Oulipian snowball, a poem based on the constraint of increasing the number of words, syllables, or letters per line. True to their methods, Oulipians also inverted the snowball (the melting snowball), and combined the two types, producing expanding and contracting lines, and hourglass-shaped poems with a visual and rhetorical shape in the same tradition as George Herbert’s “Easter Wings.”

Rettberg describes three kinds of snowball poems— by word, by syllable, and by character— and does the groundwork for all three in the Ruby program, but chose to publish only the word and character snowball poems, leaving the syllable counts to inform an original variation on the concept: the doubling. Let’s briefly examine all three to see how their rhetoric unfolds.

ALSO BE WELL

any how
them or us
air water and land
back where we come from
when will we be through with this
it could come from water or the air
he will be old and with time will follow
way to make her see me for what I am
at the time he could still learn a thing or two
at this place we would try to make a go of it

This random sampling from the snowball by word only has 10 lines, which build a strong sense of progression without straining the reader’s train of thought. The thematic focus on human relationships is strengthened by the use of a speaker addressing his/her partner. Different from the tanka, which had syllable constraints that kept their lines short, here we get the benefit of seeing how more complex relationship issues find expression in longer lines.

The random sample of a snowball by character has been quoted via an image because the style sheet for this blog would change the font from Courier to a variable width font, ruining its visual effect.

A whopping 36 lines in length, this poem builds more through visual accumulation than by meaning, though it helps to note that there are very few variants in the lengthier line level. It’s as if the poem was building toward one of the two final lines, which makes a strong last impression for the reader to think about the poem, even if the middle text doesn’t.

The doubling poem presents increments in which the number of syllables per line doubles until reaching 8 syllables, and then it divides itself by two until it ends. It is almost like an accelerated losange snowball or one which has been processed by a lossy compression algorithm that accentuates its rhetorical urgency.

AFTER WHAT YOU SAID

even out
where is your place
because we are so different
the water came and went through us
off to try again
even if

The troubled relationships that are such a clear motif in the “Frequency” poems find powerful expression in this form, capturing a moment of crisis in this randomly sampled poem. The open ended endings in this and other poems leave things unresolved, perhaps ripe for another episodic tension in the relationship.

This series of Oulipian inspired poems expose the impact of line length in poetic expression, particularly because so many of the variables (like word choice) are controlled in this experiment.

This is part 3 of a series of 5 postings on poems generated by the “Frequency” program. For an overview of the series, visit the first entry.

The terza rima and sonnets generated by the “Frequency” program have two main challenges: coherence over a greater number of lines and the rhetorical expectations that come with the form. The easiest part of all this is rhyme, since Rettberg created 12 arrays according to line endings and up to 7 rhyming variables that draw from them to create poems with the necessary rhyme schemes..

Terza rima, as a set of three line stanzas with an ABA BCB CDC DED rhyme scheme, is therefore simple to assemble. The problem is that this poetic form creates momentum by including within each tercet the key to the next tercet— and the ear craves the completion of the rhyme. It is therefore understandable that Dante created this form for his lengthy narrative poem The Divine Comedy. Let’s see one  of Rettberg’s poems.

LIKE A MAN FROM A DIFFERENT WORLD

play as if you know
two were here before you
think back long as you can go

place it where you want to
we need a hand
great to be with you

form a new land
long was the day
from water to land

people get out of the way
why ask me
of these people what can we say

The poem is surprisingly coherent, particularly when the generator fills the rhyme with a line that ends with the same word, as is the case with the third stanza. Perhaps the result is a fortunate one that works well conceptually from the title to the last line, describing a man who wishes to live in isolation, building a land to get away from people. Perhaps we are willing to create short causal chains, particularly when the lines make such recursive use of common words.

Sonnets are trickier because while their rhyme schemes are simple to replicate, the rhetorical progression is more difficult, particularly the volta. In the Petrarchan sonnet, the first 8 lines are meant to present an idea with the 9th line offering a sharp rhetorical turn that is explored in the remaning 5 lines. A random one should provide some insight.

AWAY FROM ME

on her need would he try his kind hand
 said way out there man
 way out there man
 place the letter in my hand
and you move through the land
 sentence the boy as a man
 word as the work of man
 or we could set out for a new land
name all who you know
 like you were in me
 there will come a day
her look was such that he would not know
 answer me
 they take our people away

Does the rhyme scheme seem a bit too repetitive? I suspect the program randomly chose two similar rhymes and churned out a poem that creates an inadvertent volta just from the pleasure of getting away from the rhyme in the first eight lines, though its repetition of words does help keep the lines conceptually focused. Will this work with the Spenserian sonnet and its interlocking rhyme scheme closed by a couplet?

THINK ABOUT WHAT YOU TRY TO DO TO ME

find another to give it to you
 why ask me
 most of all I need you
 would you like to see
follow me
 word as the work of man
 thing that you can tell me
 study and learn to be a good man
word as the work of man
 read him the letter
 he will help when he can
 answer my letter
first there is a sound
has he been around

The rhyme scheme across stanzas build momentum much like terza rima, with repetitiveness creating conceptual bridges with rhymes that might make you want to jump off from them. The selection algorithm prevents accidental rhyme repetition, so the couplet at the end at least offers a refreshing rhyme variation, if not necessarily a volta.

The Shakespearean sonnet… wait a minute, Rettberg didn’t publish any in “Frequency.” But he did design the program with that famous type in mind:

• He wrote only 7 rhyme variables: just enough to to get to the Shakespearean GG couplet. And he only gets as far as using E rhymes in the generated poems.
• He wrote a subroutine to produce them, readable in the Ruby program below:

shakespearian = Proc.new do  
puts ‘———————————————————’
puts frequency[rand(200)][rand(10)].upcase
puts
puts A[rand(A.length)]
puts ’ ’ + B[rand(B.length)]
puts ’ ’ + B[rand(B.length)]
puts ’ ’ + A[rand(A.length)]
puts C[rand(C.length)]
puts ’ ’ + D[rand(D.length)]
puts ’ ’ + D[rand(D.length)]
puts ’ ’ + C[rand(C.length)]
puts E[rand(E.length)]
puts ’ ’ + F[rand(F.length)]
puts ’ ’ + F[rand(F.length)]
puts ’ ’ + E[rand(E.length)]
puts G[rand(G.length)]
puts G[rand(G.length)]
puts ‘———————————————————’
end

Why didn’t he generate them? Is it because his lines aren’t in iambic pentameter and they wouldn’t sound like a sonnet? Is the constraint of posing an idea that gets elaborated with a minor volta and a couplet that transforms the whole poem too much to leave to chance? Was the Bard’s name too much of a reminder of the richness of his vocabulary versus the “base units of language” displayed here?

Perhaps it’s better to add some sonnets into the mix, but not so many that the reader tires of them. A little sonnet goes a long way, and this experiment  seems to be more successful at shorter and simpler pieces.

This is part 2 of a series of 5 postings on poems generated by the “Frequency” program. For an overview of the series, visit the first entry.

The Japanese Tanka (which literally means “short poem”) is represented here as a 5-line poem in which the first and third lines are selected from the 5-syllable line array, and the second fourth and fifth are selected from the 7-syllable array. A title is selected randomly from among all the lines in the data set and displayed in upper case. In a short poem like this, the title is probably the most important line because it sets up the conceptual frame by which the whole poem will be interpreted. Take for example, Tanka number 30:

END OF US

house is not the same
way different than before
ask them to end it
old men just want it to end
that you will take to the air

This poem’s brief title suggests a relationship near its end, one with a past that was better but in which life together is no longer tenable. The speaker addresses the other person in the relationship, asking him or her to confer with others in their life to end the relationship and fly away. This is particularly poignant when we consider that tanka has a strong cultural association with lovers exchanging coded messages. Other tanka in this sequence provide a similarly themed and textured language with a remarkable sense of cohesion. Is it the brevity of the poems that keep their short lines somewhat related? Let’s test this against an even shorter Japanese poetic form, the haiku.

Haiku became prominent in Imagist Poetry because of their compression and focus on images. Consisting of three lines— a 7-syllable line preceded and followed by a 5-syllable line— this form juxtaposes images, along with an understated comment and a seasonal reference. Haiku were used to punctuate moments in larger poetic sequences, thought they are often read and appreciated in isolation.

Haiku are not difficult to generate based on syllable count, and non-sequiturs actually suit the form nicely because they allow readers, but the seasonal or natural reference is trickier, given that only four words in the original data set— water, air, animal, and oil— belong to nature. Knowing about the thematic constraint helps, because the mind will interpret seasonal references in the context of the form when it might not have otherwise.

DOWN ON YOU

other than this world
thing that will be good for us
day is still to come

Rettberg is no stranger to inverting a nature reference with an urban one, as he did with Tokyo Garage, but there aren’t many more urban words in the original data set. These 200 most common words (“these base units of our language” as he intuitively calls them) seem to be very close to the core of human experience: basic needs, such as relationships with other people. It is little surprise that the lines he wrote revolve around human relationships, as can be seen in the haiku above (and many others).

The word cloud at the top of this entry is a Wordle visualization of the 50 most commonly used words in Rettberg’s 2000 lines (as is the case of the variable name “frequency”— not really a part of the data set, so please disregard). This analysis highlights word choices which are the building blocks for a recurrent theme brought out by reading the tanka and haiku generated by this program.

Rettberg wrote and the machine revealed some of what he had to say through these short poems. I wonder what the longer, more complex poetic forms will reveal both about his thematic concerns and how the output of this generator fits within its traditions.

P.S. Thanks for the correction on my misuse of “sentences” instead of lines, Scott!

The “Frequency” project is an investigation into Oulipian constraint-based writing in several levels:

1. A linguistically determined lexicon of the 200 most frequently used words in the English language— an endeavor that is already built upon the constraint of the corpus that analysis is based upon: Oxford Online and Google have both famously made such analyses, with massive and different data sets. 
2. Rettberg then wrote 10 lines of verse beginning with each word, using only the words on the list, for a total of 2000 sentences. This alone is a massive creative endeavor which he describes as “painstaking work.”
3. He analyzed these lines to develop a data set, organized in arrays based on the following variables: frequency rank of initial word, rhyming end words, number of syllables per word, number of syllables per line, number of words per line, and lines per string length (number of characters, including spaces).
4. The analysis embedded in these arrays allowed Rettberg to write algorithms to generate 200 instances of 10 different kinds of formally constrained poems, some from well established poetic traditions, one developed by the OULIPO, and three he developed himself.

The version published online (linked to above) is a set of 2000 poems (200 poems for each of the 10 poetic forms) generated by the program he wrote in Ruby (used in steps 3 & 4 above). Each poem is in an HTML page with a JavaScript timer that randomly redirects the browser to a new poem or one of three informative pages about the project every 30 seconds. This lends itself to a meandering exploration of the poems, as well as for an installation of the work. I like to have a second screen hooked up to my laptop computer with “Frequency” on a fullscreen browser window, allowing me to read it from time to time.

More systematic approaches to reading it are also available, such as deleting the name of the .html file on the browser’s address bar to get a directory listing of all 200 poems in that form, or manually changing the number value to read more generated works in that style. Even allowing the work to run for a while and use the back button on your browser to produce a list you can select from is a way to read more selectively.

However you choose to read the poems generated by this project, there is one question that hasn’t been addressed yet by this posting. Are the poems any good? How successful is the result of this obsessively recursive creative investigation of constrained writing?

Over the next four entries, I will perform systematic readings of the poems produced by this project, considering how effectively they engage the poetic traditions they formally embody, organized as follows:

Short (Japanese) forms:

• tanka
• haiku

Longer (European) forms:

• Terza rima
• Petrarchan sonnet
• Spenserian sonnet

Oulipian forms:

• snowball (by character)
• snowball (by word)
• doubling (an original form based on the snowball)

Original forms:

• two towers
• foursquare

At the end of this investigation, I will return to assess the project as a whole, hopefully with answers to some of the questions posed above.

I say “hopefully,” because I am also writing under constraint— reading one group of poems and writing about it, every day, and I honestly don’t know what insights the future might have in store.

Shall we explore this together?

As the story goes, when Nick Montfort published “Taroko Gorge” in his website in early 2009, his longtime friend and creative collaborator Scott Rettberg decided he would take the structure of the work and invert it, so he changed the dataset for the variables, modified the formatting of the page, crossed out Nick Montfort’s name, inserting his own above it, and published the remix in his own website. (For a complete, detailed, and quite humorous account, read the second page of the attachment in this page). He published this statement in the source code:

This here is a total remix of the classic and elegant generated nature poem Taroko Gorge by Nick Montfort. He wrote the code here. I hacked the words to make it more about urbanity, modernity, and my idea of Tokyo, a city I have never been to. Remixed March 18, 2009 by Scott Rettberg.

A close inspection of Tokyo Garage and its source code reveals a much more extensive dataset for each variable, which producing greater variation in the generated poem. Of course there are many more variables in a city like Tokyo than one might notice while walking through Taroko Gorge. The question is whether the increased complexity of an urban environment, represented through a poetry generator designed with a simplified and organized natural system in mind, can produce compelling output.

It can and does. Montfort’s poem is meditative and animistic, personifying the rocks and crags, as they contemplate themselves and each other in a geological time frame. Rettberg’s remix is full of surprises and humor, as it comes up with absurd and sometimes violent situations. For instance, in the third stanza of the screen captured image above, we see simple subject-object relations (“Dragons assault the kids”) escalate to affect the entertainment industry, with financial (“Funds drop”) and/or eschatological consequences (“ Movie stars evacuate”). Rettberg’s poem generates a complex media ecosystem, in which a small creative act in a garage in Tokyo can have a global impact. The storage function of a garage filled with memorabilia allows for the world to be represented within and the capacity for endless play with all these wonderful toys echo the code and its execution.

It is remarkable that such a simple construct can be successfully repurposed to create such distinct works, which is why J.R. Carpenter and many others (myself included) walked through the door that Scott opened and started to refurbish the place.

Over the next few weeks, this blog will be exploring the Taroko Gorge remixes, one per day, seeing how they extend the code’s capacity for meaningful poetry generation.

Upon completing my reading of the poetry in the Electronic Literature Collection, Volume 1 (tagged as ELC1 in this blog), some numbers and reflection on the works within is in order.

The ELC1 contains 60 works of e-literature, 38 of which are classifiable as poetry. Some of these works are certainly on the margins of what one might consider poetry, but I chose to include rather than exclude when in doubt. The authoring software distribution for these works is as follows:

The dominance of Flash as an authoring software and mode of publication is clear during the time this Collection was published (2006), perhaps for its smooth integration of multiple media, smooth animation, precise scheduling capability, and its ability to imbue a work with responsiveness. Mutability is not its strong suit, except when linked to user interaction. The poets who sought to create strongly combinatorial or generative works went with scripting languages such as Java, JavaScript, DHTML, and Processing, perhaps because they allowed more granular control over the programming codes.

The textual behaviors exhibited by the works in the collections are distributed as follows:

Of the 38 works, the most used behavior was scheduling (25), indicating an interest in exploring the possibilities of writing in a time-based medium. This was followed closely by responsiveness (24), which suggests that the other fascination lay in including the readers in the performance of the work in ways that the work could respond to. Static and kinetic texts were equally represented with 20 poems apiece, which might mean that they hold an equal interest for poets. The relatively low number of mutable poems (15) may be linked to the prevalence of Flash as authoring software, but not the 17 aural works, which is easy to handle with that program. Still almost half of the works (45%) incorporated some sort of sound, which points at some interest.

These numbers are from a relatively small sample, and one chosen by an editorial team that perhaps sought to represent a variety of practices in e-literature , so we cannot attribute too much importance. But they do point towards a trend in the development of e-literature, one that may change or become more defined over time.

This Collection was very well reviewed and received by the academic community, and with good reason, since the editorial board is composed of recognized poets and scholars in the field: Katherine Hayles, Nick Montfort, Scott Rettberg, and Stephanie Strickland. For a thorough review, read “Letters that Matter” by John Zuern.

I started this blog with this collection because I knew it was full of quality work, and I’m grateful to the poets that crafted these experiences and to the editors who chose them and curated them so well. Now onward to the Electronic Literature Collection, Volume 2!