2 min read

Google's Poetry Translation Falls Short

Human translators remain necessary, for the time being.
Technological progress is often thought to come at the expense of humanity. When the economy revolved around physical labor in the 19th century, there was fear of mechanization. American folk hero John Henry is said to have died after winning a race pounding railroad spikes against a steam hammer.

In the knowledge economy, John Henry's role has been played by the likes of Russian chess champion Gary Kasparov, who battled against IBM's Deep Blue and lost in 1997.

Google's rise has stirred similar anxieties about human obsolescence. Search for "Google is" and Google itself will suggest "Google is Skynet," a reference to the malevolent computer consciousness from the Terminator films.

Already a focus for societal worries about progress, the company's effort to develop a computer system that can translate poetry across languages may provoke further unease, particularly among writers, a group that while not particularly well-compensated has at least been spared the fear of being replaced by a machine. Sure, there's already assembly-line writing from the likes of Demand Media, but high-quality text and translation still requires human intelligence.

Whether that will always be the case may depend on the success or failure of efforts like Google's attempt to apply machine translation to poetry.

In a recent blog post, Google software engineer Dmitriy Genzel presents an update on the company's poetry translation research, work that's being presented this week at the 2010 Conference on Empirical Methods in Natural Language Processing (EMNLP).

"Translating poetry is a very hard task even for humans, and is clearly beyond the capability of current machine translation systems," wrote Genzel, who co-authored the paper being presented with Jakob Uszkoreit and Franz Och. "We therefore, out of academic curiosity, set about testing the limits of translating poetry and were pleasantly surprised with the results!"

Google's researchers have succeeded, but only so far. They've managed to build a system that can translate a poem and retain the rhyming scheme and meter or can transform it into a new form -- from limerick to haiku, for example.

But that transformation comes at a cost: The meaning gets lost in translation.

"It seems that at the present state of machine translation, one does indeed have to choose between getting either the form or the meaning right," Google's research paper states.

Human translators can still do better, but Google hasn't given up yet.