Machine Poetics

Collaborations with Google Translate in transforming a poem.

June 6, 2018 - 5 minute read -

The language we use shapes the world we see. The number of words we have to describe a particular color will influence how many shades of that color we’re able to perceive; conventions of language such as direction or agency may even dictate our view of events or problems.

In some ways, it’s ridiculous to think that a computer could ever translate a text with the same acuity as a human — what happens to the emotional nuances, the varied hues of meaning that emerge from choosing one specific phrasing over another? Machine translation is pretty decent at getting the point across for many languages, but it’s often still clunky, the crudest bones of meaning as seen by a computer.

Recently, however, I discovered an interesting application of machine translation. Rather than using a thesaurus to find alternative word choices for a poem I’d written, I ran it through Google Translate several times, going from English to Chinese to Latvian to Welsh and so on, converting the same text back to English periodically and seeing what came out. It turns out this back-and-forth translation isn’t idempotent, resulting in some really interesting mutations over time.

This was far from a scientific process, but here are some observations I made along the way:

  • The word “sweater” got turned into “jumper,” specifically after the translation to Welsh. :)
  • The title, “cross streets”, didn’t last for very long, turning immediately into “intersection” and then into “crossroads.” At some point, it became “join,” and only diverged further from there.
  • One of the more intriguing transformations was from “awake while the mist / swallows the city” to “I woke up to the fog. / The city sweats.” I never would have come up with this image — the city sweating — on my own, but I love how much life these three short words contain, and how radically this changes the image.
  • (Another thing to note is the separation of these lines into two sentences, whereas the original was one phrase split across two lines…it seems like Google Translate imposes its own form of sentence structure, independently of original punctuation and line breaks.)
  • Some of the phrases went through complete reversals of meaning; in the last two iterations I did and kept, “The city is really strong” became “The city is very weak”. I’m actually quite curious as to why this would happen…

Here is the series of poems, starting with the original and proceeding through increasing numbers of machine translation “layers.” Sadly, I neglected to keep track of the exact languages and number of transformations between each poem, but I’ve noted down the approximate changes between each return of the text to English. Perhaps the final effect is gimmicky; I think it does become hilariously stilted towards the end. Still, I really enjoy reading the whole sequence and watching meaning melt and warp as it goes on. In each iteration, I find something new and unexpected that changes what I take away from the poem; even the tiniest alternation is interesting (an added punctuation mark, a change from “it” to “I”). Even the most garbled versions have some root in truth, and together they paint a picture far beyond what I originally imagined. I like to think of them as alternate universe versions of the poem I started with.

#0 - cross streets

(original text)

tell me what it’s like to be
light but not leaving,
going but not gone
awake while the mist

swallows the city and

rain colors the streets
where i find you
in every passing headlight
or always just

around the corner. talk to me

about your favorite painting
then spill the coffee
from my hands,
remind me that

you’ve been here all along

wearing a sweater
meant for someone else.

#1 - Intersection

(went through Hungarian & Spanish translations)

Please tell me how to feel
Certainly, you can not go out,
I will go away, but he was not here
I woke up to the fog.

The city sweats

Colorful rainy day road
Where can I find it?
Under each reflector
Or forever

Not far from the corner. I speak

About your favorite artists
Then, let go of coffee
From my hands,
I remember this.

Always there for you

Wearing a sweater
Give to others.

#2 - crossroads


Tell me, how to feel
Of course, you can not go out,
I’ll leave, but it was not here.
Wake up to the fog.

The city sweats

Colorful rainy way
Where can I find it?
Under each reflector
Or forever

Not far from the corner. I speak

About your favorite artists
Then release the coffee
From my hands,
I remember that.

Always in your service

Wear a jumper
Give others.

#3 - Included

(Latvian, Thai, Romanian, others that I can’t remember)

Tell me your feelings
You can not really go out
I’m going, but I was not there.
Raise his wheat

The city is strong

Colorful rainbow
Where can I find it
Under each illustration
Or forever

Not far from the corner I speak

About your favorite artist
Then get out of the caravan.
In my hand,
I remember that

Always at your service

Use a player
Give to others

#4 - Add it

(Pashto, Dutch, Esperanto, Hebrew, Yoruba…)

Tell me my feelings
You can not leave
I will go, but I do not have it
Select the cache

The city is really strong

Colorful film
I can find where
A photo
Or permanently

I talked about his ears

About your favorite artist
Then he left the road.
On hand
I miss it

This is always here

Use the player
Give to others

#5 - Add full

(can’t remember)

Tell me what I think
You can not go
I’m going, but not for me
Choose the box

The city is very weak

Show image
I can find something else
Or no

I talked about his ears

About the capabilities you want
Then he left the door.
I will not die

This is also there

Use the player
For us