I ♥ E-Poetry

This suite of four poems based on W. C. Handy’s “St. Louis Blues” was written using GTR Language Workbench— a kind of textual Photoshop that allows users to algorithmically select and transform a text. This free and downloadable Mac & Windows software tool created by Andrew Klobucar and David Ayre can be used to analyze and transform texts, generating new ones using new and historical algorithmic methods, such as the Oulipian N+7. It also allows writers to create new algorithms or sequences of transformations to act upon texts, as seen in its tutorial videos (see the Processors and Mixed Processors tutorials in the program’s Help section).

Scholar, poet, and musician Chris Funkhouser is well versed in the use of this software, listing 5 creative works produced with it in his website. These three poems were selected from his month-long participation in the WordXWord 30/30 Poetry Challenge starting on April 1, 2011. For 30 days, he used this software to (in his words) “canniballize” the lyrics to Handy’s song to generate daily poems, perhaps also inspired by the prompts offered. Funkhouser’s algorithmic interventions can be understood as a series of jam sessions or musical remixes that act upon the original the textual recording (that is, the poem)

Funkhouser’s prefaces his four poems with Handy’s lyrics, which allow us to appreciate the connection and distance from the original. This is enhanced by the repetitive structure of blues lyrics, as can be seen in the first stanza:

Original “St. Louis Blues:”

I hate to see that evening sun go down
I hate to see that evening sun go down
‘Cause, my baby, he’s gone left this town.

“St. Louis Blues 2011:”

“Removal Us:”

Note how instead of repeating the line he chose to transform the lines individually, perhaps with different algorithms, yet keeping the last word intact or rhyming to retain some of the original structure. Some of the versions are quite different from the “seed” language, yet the variations are remarkably coherent because they’re guided by Handy and Funkhouser’s creative visions and to a certain extent Klobucar and Ayre’s vision for the software tool. Keep this in mind when reading the poems, especially in the context of the possible transformations one can put language through with this software.

This tool becomes a writing instrument in Funkhouser’s expert hands, who without typing is able create conceptual poems in a compositional process one could call language jazz.

Note: The software currently works only in Macintosh and 32-bit Windows operating systems.

This work of generative Internet art presents an essay to readers that reads like an essay written by a graduate student that has done nothing but read Postmodern theory for years. The result might be brilliant, nonsensical— perhaps both— but it exists on a different reality as the rest of the world’s and is likely to have little impact on anything. You might as well pump all that high theory into a machine and put together a little program to produce some semi-random output from that lexicon and then see if readers will read the results at face value.

For this piece to have any function at all, requires a mind that is eager to project meaning onto experience. If we expect an experience to be meaningless, our minds certainly do not bother to piece together the chaos of clues that make the world comprehensible. With Chomsky’s famous pseudo-sentence “Colorless green ideas sleep furiously.” for example, we undergo an initial attempt to identify a meaningful message. Convincing the mind to choose at the crossroads between potential comprehensibility and inevitable noise is an important task.

This is a crossroads readers of poetry reach when they come across a particularly challenging poem. In many cases, they will make the interpretive leaps because they hope the poet’s intentional hand (whether real or imagined) will catch them and lend them support in their interpretation— either that or because their professor asked them to and they know they have him or her as a safety net. But what is the point in reading the output of this “behavioral art?”

Part of it is figuring out the intention behind the algorithm. What point does JudsoN want to make with a work that generates endless theory papers, with keywords, sections, figures, captions, references, and a title? Is this work a critique of this kind of writing?

But more interesting (to me) is to examine the artistry in the program— one we can’t access but we can intuit, and even reverse-engineer. And the only way we can do so is by reading multiple iterations, carefully, to find patterns, repetition, variations, places where the algorithm fails or suceeds in producing grammatical or sensible sentences. This is old-school data mining: carefully focused human intelligence scrutinizing a text to find conceptual gems. Gloriously absurd or lucid phrases that excite the intellect, even if by accident.

The search for meaning that drove the New Critics teach generations of scholars and students to perform close readings of texts, was completely subverted by Post-structuralist theory, which showed that all meaning is constructed and can therefore be deconstructed. We might have come full circle with generative works like “Essay,” carefully reading that which we know is meaningless, because we might find something worth ascribing meaning to.

This program mines articles in the New York Times home page, and using a dictionary and syllable counting algorithm and a few filters, discover sentences that can be cut into the shape of a haiku. The output of this generator is vetted by NY Times journalists, who identify the best ones for publication in the Tumblr blog, after generating background art based on the first line of the haiku. Read Harris’ about page for more details on this breakthrough in generative poetry.

Readers familiar with this blog or with the long tradition of generative poetry may wonder why I describe this project as a “breakthrough,” since electronic poetry has had a long acquaintance with the generation and discovery of haiku. Haiku generators have been around at least as early as 1967, as documented in John Morris’ “How to Write Poems with a Computer” (thanks to Adam Parrish for the lead).

Some of the earlier ones explored in this blog are “Free Haiku!” (2002) and “Exquisite Corpse Poems” (1996) both of which use relatively simple mechanisms to produce haiku, yet produce results that employ and challenge the tradition. Generation alone is a simple enough feat, evidenced by the 28,600 hits yielded in a Google search for “haiku generator.” What makes haiku generation interesting is the complexity of its algorithm and the quality of its data set. For example, Scott Rettberg’s “Frequency” (2011) generates compelling haiku because they are selected from a set of lines he wrote guided by carefully designed constraints. The time of the individually authored data set seems to be passing, as seen in a contemporary generator, Nanette Wylde’s “HaikU” (2011), which allows its audience to contribute lines to its database.

Databases are among the most powerful tools placed at our disposal in the contemporary Web, including sophisticated dictionaries and massive and ever growing data repositories, such as Twitter and The New York Times. Two automated haiku finding Twitter bots recently reviewed in this blog are “Tweet Haikus” (2013) by Brandon Wood and “HaikuD2” (2012) by John Burger. Wood has published his source code and explained how to create a similar bot, if anyone is interested, though it is not as refined an instrument as Burger’s earlier bot, which also gives back to the original authors of the tweets. Both of these credit the original source, as does “Times Haiku.” As may be evident from this brief survey, there is nothing particularly groundbreaking in this project, from a technical standpoint.

From a cultural standpoint, this is an important moment for generative literature, (which has a rich history and corpus of its own) because it has begun to be recognized and produced in mainstream sources. For the flagship of American journalism The New York Times to legitimize this kind of generative literature by producing and publishing its results, even when hedged by an essay as humble as Harris’, is a sign that electronic literature is entering public consciousness. It is now time to remind the public that literary experimentation in digital media wasn’t invented today, and that they should learn about the body of work that preceded it.

Is it prophetic for this news to break on the week of the first Electronic Literature Showcase at the Library of Congress with an exhibition curated by Dene Grigar and Kathi Inman Berens titled “Electronic Literature and its Emergent Forms?” I think so.

Electronic literature has arrived.

This Twitter character came to life in the “Reality: Being @spenserpratt” netprov, was christened “Tempspence” by Pratt’s followers (as a “temporary” Spencer), and lives on in this Twitter account, along with a community called The Tempspence poets. Their symbiotic existence was sustained by social media interactions of a group of people that came together through this netprov, and extended the life of the performance beyond its metaphorical covers.

When “Reality: Being @spencerpratt” ended and everything was revealed, Mark Marino and Rob Wittig did the Twitter equivalent of stepping from behind the curtain to bow and thank the audience, polling them for some of their favorite poetic constraints. The enthusiasm and pleasure in the interactions launched the Tempspence Poets and the poetry games continued in earnest for a while, with @Tempspence as moderator and communication bridge, but it has slowed down almost to a standstill. As participation waned, the authors seem to have concluded the story arc offering some closure by resolving Tempspence’s love dilemma even working in a crossover with “#sootfall.” For now, the character remains in Twitter, occasionally retweeting or participating in a conversation or two, perhaps awaiting a new storyline for him to return to activity.

This is at the heart of a networked improv fiction: a character existing beyond its scripted existence. Literature, film, television, and other media have many examples of characters appearing in multiple works and spinoffs abound, but this is an unforeseen improvisational extension of a performance of a character beyond the edges of a conceptualized work of fiction. Perhaps this could be framed as a coda or an encore to “Reality.” Perhaps this will be compiled, archived, and assigned a title as a kind of authorial retroactive continuity.

Note (March 30, 2013): I received a comment from Mark Marino that sheds some light on the character’s expanded existence:

I should mention that the Tempspence poets collectively run the @tempspence account with Rob Wittig & I. He’s like the Dread Pirate Roberts or the Pope, except rather than turning the name and role to one successor, we handed it over to the a group of bandits (fandits?) or church ladies, as the case may be.

Tempspence & the Tempspence Poets have evolved from symbiotic separate entities to a fused character, humming with potential. Let’s see what happens to the character in the near future.

In the meantime, the Web retains traces of their interactions, captured in Storify lists, Tumblr sites, and the short-lived Twitter archives. Here is a list of the games, as reported by Marino and Wittig in this article:

• #prattplus7: replace each noun in a famous Spencer Pratt quote with the noun 7 later in the dictionary.
• #twouplets: (Twitter couplets) rhyme your tweet with someone else’s.
• #centode: Tweet about your boyfriend or girlfriend for a collective poem.
• #shibboleth: type a tweet or post a doodle that people could use to prove it’s you.
• #ekphrastic: describe yourself in a revealing picture
• #imspencer: type a line you want to hear coming out of @spencerpratt’s mouth
• #myccb: describe the people you live with as though they were the CBB housemates
• #jungbro: describe your personality as made up of CBB participants

Read this Tumblr for more elaborate descriptions. It’s worth noting thatThe Tempspence poets continued with other poetry-generating constraints (some documented here), such as: #randompoem, Twitter Chain (“To play Twitter Chain #twain, help tell a new tale from Tempspence’s life. Each person contributes 1 sentence at a time…”), #saga, and who knows what else they’ll come up with.

This bot is a stand-in for Kazemi at the Game Developer’s Conference happening at the time of this posting in San Francisco, because he will not be able to attend for the first time in 10 years. So instead of pining away on Twitter as #GDC tweets flood his stream, he created a bot so his friends could have the pleasure of his company in their own streams, which as we know, is almost as good as his being there. If that were all this piece was, it would be little more than a Kazemi-themed Twitter equivalent of this:

But there’s more to it. Kazemi has been very careful in his design. For instance, he is up-front about it being a bot, so he isn’t really trying to fool anyone into unwittingly becoming the subject of a Turing test. He could’ve patched his bot to his Twitter account, after all, and used the #GDC hashtag, but that would violate his notion of “basic bot etiquette” (a three laws of robotics for social media bots). This bot generates tweets from a template with variables for company names, bars, games, and Twitter friends that have opted in. And that is a key component.

Reading the tweets reveals that:

• Kazemi knows GDC very well from attending it for the past 10 years.
• The data set for the bot is informed by titles of real sessions at the GDC.
• Its scheduling is fine tuned to post at irregular times during dates and times relevant to the GDC activities and consistent with a very active social life that includes board games with friends and plenty of alcohol consumption.
• This bot LOVES the GDC, and is practically a cheerleader for the event by weaving in different iterations of GDC (without actually using the hashtag) along with “colorful metaphors.”
• The total effect is very consistent with the kinds of tweets generated by the event, as evidenced below.

What Kazemi has created is a realistic performance of himself at this conference, one that can serve as an amusing way to fill that Darius-sized empty spot at GDC. It is a kind of placeholder, a reminder that he’s around, and wishes he could be at GDC, which might be just the reminder needed to keep him in the loop when something relevant to him comes up. It is also a kind of practical in-joke, since people who follow the bot and opt in to be included in occasional mentions will attract others, who might unknowingly follow the bot and even ask it questions (as has already been the case). Perhaps most importantly, this bot seems to be a critique at how vacuous the GDC Twitter stream seems to be— people might as well be bots.

Is Kazemi attempting to elevate the conversation with a parodic performance of what not to say or do?

Whatever the answer might be, @Darius_at_GDC has just woken up, after just a few hours of sleep.

And he seems to have a full couple of days ahead of him. Have a good one, Darius at GDC!

These two bots generate short template based sentences and publish them on Twitter every 10 minutes. With them Schneider demonstrates some of the versatility of the same kind of device when applied to different topics.

His first bot, “@tonightiate,” uses a relatively simple template that produces an obsessive litany of consumption.

The opening constant phrase “Tonight I ate” reads like a musical refrain, particularly when read as a focused stream on the account page. The use of tonight evokes the semantic frame of supper through the time of eating reinforces it with variations of side dishes and garnishes. By pulling in random nouns as things to be eaten, the work creates delectable logical clashes, mixing semantic frameworks with the power of pure randomness. The opening phrase adjusts to the time of day, based on Eastern Time zone, which enhances its absurdity in the context of a global social network. The bot therefore serves tweet-sized portions of meaningful nonsense that comment on how everything can be consumed these days.

Note: Schneider informed me that this poem was inspired by Tao Lin’s poem “i went fishing with my family when i was five,” available as a video of a live performance and a text version.

“@MassageMcLuhan” focuses on a smaller but richer dataset: quotes and slogans from media theorist Marshall McLuhan.

Schneider’s artist’s statement (linked to in the title of this entry), offers the source code in addition to this description.

I created @massagemcluhan, a bot that would “massage” McLuhan’s quotes—work them over completely, as McLuhan would say. I’ve noticed McLuhan’s penchant for reworking and revisiting phrases (“the medium is the message” and “the medium is the massage” being the most famous), and thought it would be interesting to rework some of these phrases by substituting various nouns into them.

And it does. Every ten minutes, an alternate reality McLuhan tweets a slogan that can vary from nonsensical to prophetic, and everything in between. Schneider’s artist statement offers close readings of some of the output, placing the McLuhan we know with one we recognize.

And he’s right. We generally don’t notice a communication medium except when it malfunctions (and spoons rarely do). And even if there was any “content” in that spoon, I’m certain @tonightiate would’ve eaten it already.

This bot mines the Twitter stream for phrases starting with “when,” extracts the clauses, and joins each phrase with a randomly selected animated GIF in a Tumblr. Here’s a more detailed description from Parrish’s blog:

A “#whatshouldwecallme-style tumblr” is one in which animated GIFs are paired with a title expressing a circumstance or mood—usually a clause beginning with “when.” I wrote a Python script to make these kinds of posts automatically. Here’s what it does:

(1) Search Twitter for tweets containing the word “when.”
(2) Extract the “when” clause from such tweets.
(3) Use Pattern to identify “when” clauses with suitable syntax (i.e., clauses in which a subject directly follows “when”; plus some other heuristic fudging)
(4) Post the “when” clause as the title of a tumblr post, along with an animated GIF randomly chosen from the imgur gallery.

This is both a critique and homage of the #whatshouldwecallme tumblr and the meme it inspired. Memes are powerfully infectious prompts for creativity, and they are particularly interesting (from a poetic perspective) when they lead to constraint-based experimentation with language.

Parrish’s bot is a codification of the meme, seeking to represent the idea at its core— that it is basically a subset of the captioning meme with “when” phrases as captions. By automating their creation through mining two social media platforms for text and images, he critiques how minimally creative and ultimately tiresome this fad can be— though the original creator of #whatshouldwecallme does an admirable job of keeping the performance going strong and fresh. But Parrish also pushes it into the conceptual, uncreatively producing non-sequiturs that our brains will try to sequitur.

This bot data mines a 1% sample of the public Twitter stream to identify tweets that could be considered haiku. It then republishes the result, formatting it as can be seen above, and retweets the original in its Twitter account. The page the haikus are published in uses random background images of nature, a nod towards the seasonal reference so valued in this poetic tradition.

This project is very similar to John Burger’s HaikuD2: they’re both bots that mine, filter, reformat and publish found poetry conceptually repackaged in the Haiku tradition. Their proximity invites comparison of the choices made in selecting and shaping the processed output. A noteworthy difference is the division into lines (in HaikuD2) versus using div tags to separate each 5 or 7 syllable cluster, particularly with how they focus attention on ideas, phrases, images, emphasizing or deemphasizing enjambment. Another is in how they handle attribution in their publication. Wood displays the original tweet in the “Tweet Haikus” page and linking back to it, but the Twitter manifestation is simply a retweet from ‏ @tweethaikuscom, without even a link back to the page generated from it or a hashtag (as Burger’s bot does) for the reader to realize that they have inadvertently written something that could be read as a haiku. More comparisons could be made, and even evaluations of their relative success, but suffice it to say that both bots are effective at what they do, and both have room to develop in order to refine their output (which both programmers have publicly expressed).

My favorite aspect about Wood’s creation is that its source code is published online, along with tutorials that lead up to its creation. Reading these excellent entries on Twitter data mining and natural language detection, one can see this project as an example and proof-of-concept rather than a completed work. In this sense, “Tweet Haikus” is a wonderful invitation to remix, develop, and build upon this published framework.

And what else might we be able to find, remix, and repackage with similar methods?

This cleverly named bot finds haiku in the twitterverse and republishes them in a recognizable format. The program “runs on @johndburger’s laptop” and even though the code isn’t available, the basic procedure can be inferred from the results as a set of steps:

1. The program uses Twitter API to pull tweets to analyze, filtering out anything that isn’t in English.
2. It uses some sort of library, like the Wordnik API to identify and count the number of syllables in all the words obtaining a total for the tweet. With this procedure, it can identify tweets with exactly 17 syllables.
3. It then determines which of those tweets can be divided into three lines of 5, 7, and 5 syllables without cutting into any words.
4. It formats the results to add: line breaks, a ” •” symbol at the end of the first two lines (to signal line breaks for Twitter clients that don’t support them), attribution to the writer of the original tweet, and the #haiku hashtag.
5. Burger then selects the best haiku or simply posts the raw results (I’m not sure), and manually post or schedules about 6 tweets per day with a 4-5 hour interval between them.

Twitter has long been a friendly environment for haiku, which fits comfortably within its 140 character constraint, but never more than now that it has added line break functionality into its Web client. The sheer amount of haiku written, found, and generated into Twitter is astounding, but there’s something special about this bot: its social dimension.

Imagine the delight at discovering that something you tweeted was found to be a haiku. Wouldn’t this cause you to reexamine your language through the conceptual frame of this poetic form? Could this encourage some to then purposefully write a haiku or two (or ten)? Similar to Pentametron, this bot draws attention to language, line breaks, word choices, surprising the writer with line-based juxtapositions, and accidental seasonal references. This is a case of found poetry, using a mechanical process that can always be refined, but already produces wonderful results.

Of course, if you’re a purist and find joy only in carefully crafted lines that follow the strict haiku tradition, then these are not the haiku you’re looking for.