There is an anecdote in the Arabic tradition about Abu Nuwas, an 8th century Arab poet of who studied poetry under Khalaf al-Ahmar.
Early in his career, Abu Nuwas requested Khalaf’s permission to write poetry. However, al-Ahmar refused to grant it until Abu Nuwas memorised 1,000 pieces of classical Arabic poetry. After some time, Abu Nuwas returned and recited all 1,000 poems to al-Ahmar. But, in a plot twist, al-Ahmar insisted that Abu Nuwas forget all 1,000 poems before allowing him to write his own poetry.
This tale crossed my mind while I was engaging with ChatGPT recently. I selected a set of vocabulary in advance and asked ChatGPT to compose a poem using it, first in English and then in Arabic. Despite frequent news and reports about the software’s errors in math and science, it was able to “produce” a poem.
Both (the Arabic and the English) poems, composed of several stanzas, followed the rhymed structure of an English poem (the Arabic version appears closer to a translation than an original work due to the lack of an established model), but it wasn’t as bad as other poems I’ve seen on social media.
However, the question remains: Can this work be classified as poetry?
The answer is a resound “no.”
I will elaborate on my answer later, but first I will go off-topic.
Twin assignments
After I asked ChatGPT to write an Arabic poem using the words “night, sadness, morning, joy, home, and forest,” the software became clearly confused. Despite the fact that I chose these words at random with the intent of creating a contradictory theme, the software froze after the first words, which translate to “O night of sorrow, my...”.
I changed the command to “compose a poem incorporating the words: Home, love, joy, light, and sky.” My first attempt lasted over 10 minutes, but on my second attempt the software quickly produced a poem titled “The Heavenly Love Poem.”
The poem, which could possibly be the software's first Arabic poems, was divided into four stanzas.
It could be considered acceptable for its attempt at simplicity, style, and rhyme, but the Arabic words and expressions conveyed no meaning or poetic imagery, and rhyming words were used out of context or misspelled — in what could be considered as necessary errors.
In summary, the outcome is similar to that of a primitive robot that is still learning how to spell and has no prior experience with the language of poetry.
On the other hand, the English poem I requested appeared to have fewer noticeable issues when compared to its Arabic equivalent. The software titled it, “A Poem of Love on a Rainy Night” and goes like this:
Love and sadness mix like rain on leaves,
Falling softly in the night like whispered grieves.
Trees sway in the wind, their branches low,
A sight to behold, like the ebb and flow.
The table stands still, a silent witness,
To the emotions that spill, like an endless bliss.
With each drop of rain, love and sadness unite,
In a dance that’s both pain and pure delight.
The night is dark, but the rain brings light,
Cleansing the soul with its misty might.
Trees stand tall, like sentinels of the heart,
Guiding us through the storms that tear us apart.
And so, the rain keeps falling, night after night,
A reminder of love that’s both beautiful and bright.
The table remains steadfast, a symbol of hope,
That through the rain and sadness, love can cope.
This poem, derived from a vast collection of poems fed into the software, shares the same scholastic characteristics as the previous Arabic example, in the sense that it relies on pre-existing, familiar concepts, but is more coherent and lacks the obvious problems found in the Arabic version.
Language challenge
The clear disparity between the two machine-generated poems is probably caused by Arabic’s overall deficiency in its connection with the Internet and machine translation. The Arabic poem exhibits scholastic naivety, straightforwardness, lack of imagination, and an inability to produce the smooth rhythm present in the English poem.
Unlike various European languages and English, Arabic has not made significant strides in the development and refinement of machine translation, and still faces significant shortcomings and loopholes. This is due to the absence of companies that undertake significant projects in this field and the lack of partnerships with international developers like Google.
