Google Tackles Poem Machine Translation with Surprising Results

Translation is a big challenge for computers. Algorithms have gotten better and systems like Google's Translate do a fairly good job, by relying on a huge amounts of data. But the translation is most of the times "usable" and nothing more.

Not to be discouraged by this, Google engineers have set out to translate poetry, to see just how capable machine translation technologies have become.

The results, Google says, have been quite encouraging, granted, based on the modest expectations that everyone had.

"Robert Frost once said, 'Poetry is what gets lost in translation.' Translating poetry is a very hard task even for humans, and is clearly beyond the capability of current machine translation systems," Dmitriy Genzel, Software Engineer at Google, wrote.

"We therefore, out of academic curiosity, set about testing the limits of translating poetry and were pleasantly surprised with the results!," he said.

"We will present a paper on poetry translation at the EMNLP conference this year. In this paper, we investigate the purely technical challenges around generating translations with fixed rhyme and meter schemes," he added.

In the paper, Google researches show how they altered the regular Translate algorithm to ensure that the translations follow the poetic form of the original.

Google says that not much had to be sacrificed in order to get it working and only made the changes necessary for the system to place words in a specific order.

Tackling something like rhyme and meter is actually much more suited for computers that it may seem. The meter is something that an algorithm can measure and determine, as is rhyme, of course.

In fact, Google's research technology can actually 'translate' any text into poetry and you can even specify the style of the poem, like limerick or haiku.

The technology is not ready for the prime time and Google says that it's too slow to release it to the public, even in experimental form. However, there have been some rather encouraging results and the researcher lists several, like the example below, in the blog post.

"A stanza from Essai monographique sur les Dianthus des Pyrénées françaises by Edouard Timbal-Lagrave and Eugène Bucquoy, translated to English as a pair of couplets in iambic tetrameter," Genzel said.

"So here's the dear child under land,
will not reflect her beauty and
besides the Great, no alter dark,
the pure ray, fronts elected mark."