Google Translate’s Emergent Poetry
Some of you will be familiar with computer poetry, poetic compositions generated by computers using algorithms. Some of you may even be familiar with computer prose, as the book The Policeman’s Beard is Half Constructed (text here). There are lots of things to say about this. Who’s the author? Is it really poetry? And what does it say if computer poetry passes the Turing test?
Last week, I stumbled upon something new in this neighborhood, care of Google Translate. You might think this would be generated by inputting something funny (but promising if you think about it) like assembly instructions or political speeches–or even something translated into a different language, then translated back. Instead, this Google Translate poetry takes as input a single, repeated Japanese hiragana character. As you can see above, the returns are surreal and delightful. (For all of these, I’ve used ‘ke’, け.)
See here and here for more examples.
For a little background, hiragana is a syllabary, so it’s not like Chinese characters where, roughly speaking, each character is a word, and these are subsequently built up into other words. An individual hiragana character can be a word, but this is also true in English with ‘a’ and ‘i’. And, like English, the meaning of the characters are not somehow built into the meaning of any word containing them. (The meaning of ‘a’ is not built into the meaning of all words that contain that letter.)
This fact about hiragana makes the results all the more interesting. In fact, you don’t need to limit yourself to hiragana to get these outputs. At his blog Riddled, Smut Clyde uses all sorts of different repeated syllables and repeated letters to generate such poetry.
You can also get different results by switching up spacing, returns, hyphens, and so on. Here’s an example using spaces.
Philosophically speaking, it’s a little different from traditional computer-generated poetry, which often takes a mass of text as an input in order to generate something new and sometimes in a similar style. Here, the program (Google Translate) is not intended to really generate anything. It’s meant to convert some existing meanings into roughly synonymous existing meanings. (It would actually be contrary to the goal if new meanings were created.) But what we see in the above examples is meaning that just sort of emerges out of language goo. It’s as if we’d shaken a tree and its twigs and leaves fell into a meaningful pattern, or if we discovered a poem floating on the top of a bowl of Alpha-Bits cereal. We might end up shaking a lot of trees to find something good, but there is a sense in which we didn’t create the thing that comes out.
I don’t have any position on this stuff. Maybe it’s a collaborative, co-authored work. Maybe we who dub the final thing ‘poetry’ are the real authors. Or maybe it’s just not poetry at all. Maybe there’s nothing philosophically controversial here. But even that would be kind of surprising, I think. In any case it’s a fun example to think about. And so much fun to play around with, too.
I’ll say one thing, after having messed around with it a little bit: As the one who enters the characters, you have some control over what comes out. You can exercise this control to varying degrees, being more hands-on (inserting spaces and punctuation, cutting and pasting, determining which lines are more or less interesting/poetic/provocative) or more hands-off (entering virtually random-length strings punctuated by occasional line returns). The more hands-on it gets, the more it feels like a collaboration, as you get inspired by what excellent random things pop out (“Welcome to the place where you can sit down with your birthday daughter”??). The more hands-off it gets, the more it feels like you’re just stumbling upon some surprisingly meaningful twigs.
Here’s a final challenge (or concession?), in the form of a one-line poem, very poetically using ‘i’ (い):
(via Language Log)