Monday, January 17, 2011

Google Poetry Translator!

Crack reporters at The Bugle, each notoriously frugal,
Thought they’d stoop to using Google for a verse-translation chore;
Past the ordinary deadline, hoping vainly for a headline
To at least avoid the breadline, where they’d eaten oft before
They were poorly paid reporters, so they’d eaten there before,
And had promised “never more”

Google searches through for matches; when it finds one, it attaches—
If it doesn’t work, dispatches it as something to ignore
Matching syllables and timing, more than this, it looks at rhyming,
Though its crude syntactic priming may yield verses you deplore
Mind you, many human poets give you verses you deplore,
Though we warn them “nevermore”

In this time-intensive screening, Google focuses on meaning,
And the finely tuned machining is the only guarantor
That computerized translation isn’t mere abomination,
And will win the acclamation of the critics by the score
Though the human poets shudder, thinking “critics by the score!
Let them visit never more!”

But a mere computer stripling, it can imitate a Kipling
Though the output may be crippling, and may leave your eardrums sore
Though it lacks a human passion, it’s a poet, in a fashion,
Not exactly Ogden Nash-ian, still it’s nothing you’d abhor
And a lot of modern poetry, the public does abhor
So they read it nevermore.

Now with German, French, or Russian verse, it’s worthy of discussion:
Will we notice repercussions from this poetry galore?
With a software package Babel fish eventually able
To re-work a verse or fable, might it save us all from war?
If we understand each other, will we still believe in war,
Or agree to “never more”?

NPR reports that Google is developing an artificial intelligence Poetry translator.
"It's what we call AI complete," says Dmitriy Genzel, a research scientist at Google. "Which means it's as difficult as anything we can attempt in artificial intelligence."

Programming a machine to simply understand language, after all, is a task IBM spent four years and millions of dollars to accomplish with its Watson computer, which competed on Jeopardy last week.

Watson understands human speech. But for a computer to understand and translate poetry, there are added problems of length, meter and rhyme.
Tell me about it.

Well, actually, they do tell us about it, in a paper on the Official Google Research Blog. And, damn them, they parodied The Raven before I did, which I'd have known if I'd clicked that one link. Oh, well. They only did one verse. But the paper is well worth reading; I've often said that what I do can be accomplished by an adequately trained monkey... looks like Google is raising the ante.


Anonymous said...

You, sir, are brilliant and much loved.

Cuttlefish said...

Yes, but poorly paid. These things have a way of evening themselves out.

Thank you, though!

Svlad Cjelli said...

I actually dislike poetry translation as a whole. I prefer a proper translation that doesn't fit into any meter.

Although those poetic translations can be good poetry in their own right, I can't help but despise them a little in their role as translations.

Cuttlefish said...

I am of multiple minds on the subject. I'd hate to miss out on, for instance, the poetry of Kafavy, which I first discovered while trying to learn Greek. Although I eventually recognized that I would never actually "get" them as a native Greek speaker would--because not merely the ideas, but also the language, the rhyme, the meter, all matter--the translated poems were much better than missing the poems altogether.

Hikmet's "Things I didn't know I loved" I have only read in translation; I have a new Turkish friend who has promised to read it for me properly, though. I have done this with Russian and Bulgarian poetry as well--read it in translation for me, and heard it in the original to hear the music in the words.

There is no ideal solution, because I have no chance of becoming native-fluent in all languages. I must say, this is a serious disadvantage to being American (thus, by definition, speaking only one language).