Bot or Not? Poetry does not compute

by hexacoto

compupoem

Passing Time
Your skin like dawn
Mine like musk

One paints the beginning
of a certain end
The other, the end of a
sure beginning

Do you think the above poem is written by a human, or generated by a computer program? That’s what site bot or not seeks to achieve: a Turing test where people try to discriminate computer-generated poems from human-written ones. By the way, the above poem, “Passing Time,” was written by a human Maya Angelou.

I have previously written about reconciling the idea of programming code as poetry, whether it’s possible to achieve “poetic beauty” in code. I posited that there might possibly be an inherent “poetic beauty” in poetry that we recognise when we decide a piece of writing is a poem. With bot or not, the idea of what makes poems poetry is taken in a different step: by identifying whether the “poems” are human or computer-generated, there must be something common with human effort that is visibly distinguishable.

That means that any piece of writing that is seen as containing enough of the essence of “poetic beauty” to be counted as a poem, can be further sub-divided into being human effort or algorithm-derived. There must be commonality between “computer poems” and “human poems” that we can decisively say, “Yes, this is the work of a human” or “This is clearly computer-generated.”

The Showcase

The site lists some of the top poems easily recognised as human or computer-written.

Poem

Generated by botpoet using JGnoetry (93% said “bot”)

Published on desserts and from pink. Symptoms
Start, 2013 as other poetry does anyone
Word in thailand and write reading one mother
Order they deserve. Well, 2013 recently released
My pants and serve throw from is a beautiful
Insane surreal once hours playing once.

Personal Space

Generated by botpoet using JGnoetry (87% said “bot”)

New poetry I might help you currently
Have been, snapping male, but it’s a rocket,
Kid man been forever idea of
My wearing a punk and brought old robot
Smog thing. Professional grown-up looking
In dusty old men remaining his tibia.

Bright

Generated by Jim Carpenter using Erica T Carter (85% said “bot”)

The name ghosts second, destroying.
The quite normal letter to the dutch throne after one year destroying the still stuck ridged snow, interpreted.
Undimmed radiance curves, primming like the column.
Lounging, as other as very high score.
Loafing a booby.
Lounging and spends.
Normal occasion joints.
Early letter gets the personal experience from the individual.
Getting, tarries however in cell.
Lounging.
Obsesses the disturbed surface.
Obsesses the abyss.

As can be seen, the above poems are mostly verbiage, and make no sense. We compare that with the poems most recognised as being written by humans.

The Fly

William Blake (87% said “human”)

Little Fly,
Thy summer's play
My thoughtless hand
Has brushed away.

Am not I
A fly like thee?
Or art not thou
A man like me?

For I dance
And drink, and sing,
Till some blind hand
Shall brush my wing.

If thought is life
And strength and breath
And the want 
Of thought is death;

Then am I
A happy fly,
If I live,
Or if I die.

Untitled

Shelby Asquith (86% said “human)

His smile was loud,
and me in my silence—
I thought it was meant
for me. Momma warned
me about boys like him.
Told me that the kinds
of boys that shined a little 
too bright might just be 
trying to distract me from 
the balled fists, the fury.
I was a fly and he lured
me straight into the light.
And oh how he burned
me, how he burned me.

O Fool

Rabindranath Tagore (84% said “human”)

O Fool, try to carry thyself upon thy own shoulders! 
O beggar, to come beg at thy own door! 

Leave all thy burdens on his hands who can bear all, 
and never look behind in regret. 

Thy desire at once puts out the light from the lamp it touches with its breath. 
It is unholy---take not thy gifts through its unclean hands. 
Accept only what is offered by sacred love.

Especially with Blake’s poem, these top poems display some things that computers can scarcely replicate. In Blake’s poem, there is a very strong meter and rhyme scheme going on. The other two read very coherently, where the ideas contained within the lines agree with each other, and a message flows from the poem to the reader. What does it take for a piece of writing/poem to resemble human effort? We look at computer-generated poems that people thought looked human.

#6

Generated by Janus Node using Janus Node (69% said “human”)

you

  are

      inscribed
          in the
           lines on the
     ceiling

      you

 are

   inscribed in
         the depths
   of
         the
    storm

A Wounded Deer Leaps Highest

Generated by Poets using Ray Kurzweil’s Cybernetic Poet (67% said “human”)

A wounded deer leaps highest,
I've heard the daffodil
I've heard the flag to-day
I've heard the hunter tell;
'Tis but the ecstasy of death,
And then the brake is almost done,
And sunrise grows so near
sunrise grows so near
That we can touch the despair and
frenzied hope of all the ages.

some men

Generated by Every Google User using Google Predictive Search (67% said “human”)

some men just want to watch the world burn 
some men just want to watch the world learn 
some men just want breakfast

In these instances, how did algorithmically-generated sequences of words suddenly gain the verisimilitude of human effort, whereas the computer-like computer-generated attempts shown above failed? Finally, we look at poems written by humans that people thought were computer generated.

Cut Opinions

Deanna Ferguson (76% said “bot”)

cut opinions tear tasteful
hungers huge ground swell
partisan have-not thought
green opinions hidden slide
hub from sprung in
weather yah
bold erect tender
perfect term transparent till
I two minute topless formed
A necessarily sorry sloppy strands
hot opinions oh like an apple
a lie, a liar kick back
filial oh well hybrid opinions happen
not stopped

Cinema Calendar Of The Abstract Heart – 09

Tristan Tzara (69% said “bot”)

the fibres give in to your starry warmth
a lamp is called green and sees
carefully stepping into a season of fever
the wind has swept the rivers' magic
and i've perforated the nerve
by the clear frozen lake
has snapped the sabre
but the dance round terrace tables
shuts in the shock of the marble shudder
new sober

Red Faces

Gertrude Stein (69% said “bot”)

Red flags the reason for pretty flags.
And ribbons.
Ribbons of flags
And wearing material
Reason for wearing material.
Give pleasure.
Can you give me the regions.
The regions and the land.
The regions and wheels.
All wheels are perfect.
Enthusiasm.

The Experiment

That we have seen the most human-like human/computer-written poems, and most computer-like human/computer-written poems, can we draw parallels for what constitutes human effort in poetry? On the technical side, can we say that, as per Blake’s poem, prosodic and auditory cues such as stress, meter, and rhyme give poems a sense of human effort, such as where by reciting “Tiger, tiger, burning bright/In the forests of the night,” we can not only hear the rhyme but feel a sense of constant rhythm to the poem?

Surely that can only be achieved by humans? Not rightly so. Assuming a bot program has access to a dictionary, and the stress, meter, and phonetics of all words contained therein all mapped out, how hard would it be to code for something that reads like human poetry? (Following is not real code, but an idea of how the code should behave)

<write human poetry>
component: alternate stress, strong-weak
component: vowels at end of line match; SET1:AAB SET2:CCB
SET1
component structure line1: [pronoun][conjunction][pronoun]
component structure line2: [verb][preposition][location]
component structure line3: [verb][noun]
SET2
component structure line1: [pronoun][verb]
component structure line2: [conjunction][verb][noun]
component structure line3: [conjunction][pronoun][verb]
And I just described:

Jack and Jill
went up the hill
to fetch a pail of water.

Jack fell down
and broke his crown
and Jill came tumbling after.

A bot could browse through a dictionary and probably come up with something similar. Granted, I “retro-wrote” the code, where I already had a poem in mind and wrote the “code” after, but if I can break “Jack and Jill” down into an algorithm that can be reproduced, using auditory and prosodic cues, then surely it is solely not that that determines human effort in poetry? However, if a program relies solely on prosodic and auditory cues, what’s to prevent it from putting in random words that fit those cues but make no sense in sequence? For example:

Bird and ball
swirled by the mall
and cocked a round of seaweed

Truth flew out
where running lout
and cops were sniffing soft beads

The prosodic and auditory cues of the above poem match “Jack and Jill” yet it makes no sense, and it is likely that people would judge it to be written by a bot. So what else is required for poems to be recognised as human effort?

The other thing you’ll notice where human-like poetry trumps computer-like poetry is coherence of ideas. In the poems that read human, most of them have ideas that agree with either a general theme, or the lines preceding and following them. The ideas contained in each line also display a progression, where there is something being explored or developed. The computer-like poems tend to show disjointedness of ideas.

Perhaps humans are predisposed to pass continuity and coherence as hallmarks of humanity.

Is coherence then unique to humans, or can computers imitate coherence as well? Let’s see if we can imitate coherence with an algorithm as well, with added prosodic and auditory cues. To achieve that, we need thematic cues. I’m going to use the following poem, “I wandered lonely as a cloud” by William Wordsworth.

I wandered lonely as a cloud
That floats on high o’er vales and hills,
When all at once I saw a crowd,
A host, of golden daffodils;
Beside the lake, beneath the trees,
Fluttering and dancing in the breeze.

<write coherent poem>
component: alternate stress, weak-strong
component: SET last line: [alternate stress, weak-strong]=false
component: vowels at end of line match; SET:ABABCC
component: [CENTRAL X(n)] designate IDEA
component: [CENTRAL X(n)]; X=verb, noun, adverb, preposition, adjective
component: [CENTRAL X(n)] must agree with THEME
component: [CENTRAL X(n)] must agree with [CENTRAL X(n±≥1)]
component: [CENTRAL X(n)] either expand or progress [CENTRAL X(n±≥1)]
SET
component structure line1: [pronoun][CENTRAL verb(1)][adjective][preposition][CENTRAL noun2]
component structure line2: [conjunction][CENTRAL verb3][adjective][preposition][CENTRAL noun4][conjunction][CENTRAL noun5]
component structure line3: [conjunction][adverb][pronoun][CENTRAL verb6][noun]
component structure line4: [noun][adjective][CENTRAL noun7]
component structure line5: [preposition][CENTRAL noun8][preposition][CENTRAL noun9]
component structure line6: [CENTRAL verb9][conjunction][CENTRAL verb10][preposition][CENTRAL noun11]
THEME:nature
(1) verb-noun agrees with (2); (2) agrees with THEME
(3) verb-noun agrees with (2);(4),(5) agrees with (2); (4),(5) agrees with THEME
(6) verb-noun agrees with (7)
(7) agrees with THEME
(7) preposition agrees with (8),(9); (8),(9) agrees with THEME
(10),(11) verb-noun agrees with (7); (10),(11) agrees with THEME
IDEA1:[(1),(2)]–>IDEA2:[(3),(4),(5)]
IDEA3:[(6)(7)]–>IDEA4:[(8),(9),(10),(11)]

Does the above make any sense? It took me a while to try to break down “I wandered lonely as a cloud” into a vague enough algorithm that in my opinion still represents the poem while hypothetically still able to reproduce another poem. Let me explain the above “code.”

The poem has various components, including a weak-strong stress meter; but the last line of the set breaks the meter. The last phonetic features of certain lines must match; in this case ABABCC. Within the poem, there are certain things, designated as [CENTRAL X] where X can be a verb, noun, preposition, adverb, or adjective. These [CENTRAL X] designate a contained IDEA, which is a sense of what that line means. The [CENTRAL X] must agree with a preset THEME, which in this case, is “nature”; where the words must be somehow relevant to “nature.” such as “hill,” “daffodil,” and “cloud” being all words related to “nature.” Not only must [CENTRAL X] agree with THEME, it also has to agree with each other, one or more preceding or following it. It does so not only grammatically, but also has to expand or progress it in a logical way.

IDEA1 contains [CENTRAL (1),(2)], which progresses into IDEA2, containing [CENTRAL (3),(4),(5)]. IDEA3 contains [CENTRAL (6),(7)] which progresses into IDEA4, containing [CENTRAL (8),(9),(10),(11)].

You know, even after so much postulating, I’m still not sure I have successfully “retro-coded” Wordsworth’s poem. Maybe it is coherence of idea that seem unique to human effort, and that humans are predisposed to finding order in nature. My head hurts from trying to break poems down like that. Maybe someone else can do this better than I can. Feel free to leave comments.

Advertisements