ROBERT SIEGEL, HOST:
Could a machine at least write a love poem, a poem moving enough to stir the human heart? Well, not yet. But here's a step in that direction.
(SOUNDBITE OF MUSIC)
MELISSA BLOCK, HOST:
So begins a sonnet written by Nathan Matias with assistance from his computer and William Shakespeare. Matias is a student at the MIT Media Lab.
SIEGEL: And Shakespeare, of course, is that famous dead English writer.
A computer program drew on a database of Shakespeare's works - only words used by the Bard.
BLOCK: Then as Matias composed the sonnet, the software offered a word that might work.
: The software suggested words that Shakespeare might likely use in that situation.
BLOCK: Matias could pick that word or another. It was his sonnet confined to authentic Shakespearean language. It's the same predictive software we see when our devices try to finish our sentences and suggest the next word.
: The software was making suggestions and guiding me to themes and words that Shakespeare would likely use.
SIEGEL: But Matias had the last word.
(SOUNDBITE OF MUSIC)
SIEGEL: Matias attempted to do this with the works of other poets. But it turned out that some modern-day poets were too unpredictable for predictive software to help much.
BLOCK: But once the predictive part is mastered, the next step would be poetry created entirely by software. Computers could bang out all sorts of grammatically correct verse.
SIEGEL: Only then would the humans come in as readers who approve or disapprove. And he suggests it could be rated by crowd-sourcing. Bad poems would be filtered out and the best ones survive.
: We may well see people creating large amounts of automated poetry and then finding out which poems are popular.
BLOCK: That's the MIT Media Lab's Nathan Matias. He expects to see a successful automated poet in his lifetime.
(SOUNDBITE OF MUSIC) Transcript provided by NPR, Copyright NPR.