Artificial eloquence: the weird science of computer-generated poetry

In yet another example of computers churning out art, Paul Thompson has built a C++ script that produces snowballs — poems where every line is a single word, with each successive word one letter longer than its predecessor. "I thought that someone must have created a program to generate these, and I was surprised that I couldn’t find one even after some pretty thorough Googling," Thompson writes on his blog. That led the part-time programmer to develop his own solution. Thompson's code scans through a massive text file of numerous combined works from Project Gutenberg. All character names are ignored as the poem is crafted, as are any foreign phrases included in the source material. His formula also relies on Markov chains to build out the finalized poem, though Thompson readily admits his somewhat elementary approach isn't perfect. "The output will, for the most part, be absolute rubbish," he says. "But there will be wheat in the chaff." We've included one of those eloquent examples below.

The newspaper of record is venturing into poetry. Today may be April Fools Day, but Times Haiku — a new side project from The New York Times — is a wholly serious endeavor launching just in time for National Poetry Month. Nieman Journalism Lab has the full rundown on the Tumblr-based project, which spits out haikus generated from sentences on the NYT's homepage. It accomplishes this by matching words up against a dictionary that helpfully includes syllable count. Selections published to the blog are curated by humans, a necessary step since its creators admit Times Haiku "can't distinguish between an elegant verse and a plodding one."

To build the whimsical project, developer Jacob Harris coded a script that digs through front page stories in search of ideal words and phrases and then arranges them in proper haiku format. Designer Heena Ko and software engineer Anjali Bhojani helped push the project across the finish line; they also decided to make haikus appear as images, a smart move that makes them easily shareable across various social networks.

One developer has found the perfect way to take advantage of Twitter's new support for line breaks: haikus. Using the public Twitter stream and a Python tool that can determine the number of syllables in most English words, Brandon at h6o6 has found a way to collect the haikus hidden in plain sight on Twitter. You probably won't see the poems collected at Tweet Haikus in the next great found poetry anthology, but that doesn't mean there isn't some accidental beauty in the results — after all, the charm of @Horse_ebooks has nothing to do with well-prepared prose.

Valentine's Day is fast approaching, which means it's time to start writing poetry and compiling mixtapes. Fortunately, Spotify's Playlist Poetry lets you kill both lovebirds with one stone. The app, announced yesterday, invites users to create poems entirely from song titles, and share the resulting playlist with their Special Someone. You can search Spotify's database for a specific word or phrase in a given title, and arrange the results into verse using a word magnet-like interface.

The emphasis here is on words, rather than music. You can't search by artist, album, or any field other than song title, which can result in some unintentionally strange or nonsensical playlists. It's not exactly a new concept, but Playlist Poetry certainly makes it easier to visualize and construct your words of affection — or antipathy, as the case may be.

If you're a Twitter user, you've likely come in contact with the spam account @Horse_ebooks at least once or twice, probably when someone you follow discovers him and loses all self control, retweeting every one of his many gems. If, however, you're unfamiliar with the Twittering world's most beloved equine chatterbox, here's a quick summary: @Horse_ebooks is a spambot, created to promote and presumably sell ebooks (Gawker investigated his origins and creator in depth earlier this year). Sometime yesterday or the day before, @Horse_ebooks passed the 100,000 follower mark: a remarkable achievement for a human, but a somewhat stunning one for a bot.

The account Tweets out seemingly random phrases, from seemingly horse-related ebooks, resulting in verbose poetry such as: