





































Humanist Studies & the Digital Age, 7.1 (2022) 

ISSN: 2158-3846 (online) 

http://journals.oregondigital.org/hsda/ 
DOI: 10.5399/uo/hsda/7.1.6 

 
 

 

  1 

Poetry at the first steps of Artificial Intelligence 
 

Christina Linardaki 

Abstract: This paper is about Artificial Intelligence (AI) attempts at writing poetry, usually 

referred to with the term “poetry generation”. Poetry generation started in the Digital 

Humanities, which developed out of humanities computing; nowadays, however, it is part of 

Computational Creativity, a field that tackles several areas of art and science. 

No matter which field is undertaking the venture of poetry generation, it is certainly a 

fascinating process, as it essentially involves a reinvention of how the human brain works. 

Moreover, if the vision is to arrive at a novel form or idiom of poetry, the product can be 

altogether refreshing. But so far, AI attempts at poetry generation are not so different from 

equivalent human attempts based on algorithmic processes. This means that, in a way, AI 

attempts have dovetailed with poetry’s long tradition as an art form in evolution.  

In the paper, it is first examined why poetry was chosen among other literary genres as a field 

for experimentation. Mention is made to the characteristics of poetry (namely arbitrariness 

and absurdity) that make it fertile ground for such endeavors and also to various text- and 

reader-centered literary approaches that favored experimentation even by human poets. 

Then, a rough historic look at poetry generation is attempted. This is followed by a review of 

the methods employed — either for fun or as academic projects — along Lamb et al.’s 

taxonomy (“A Taxonomy of Generative Poetry Techniques”), which distinguishes between 

mere poetry generation and result enhancement. Another taxonomy by Gonçalo Oliveira (“A 

Survey on Intelligent Poetry Generation: Languages, Features, Techniques, Reutilisation and 

Evaluation”), dividing between form and content issues in poetry generation, is also briefly 

presented.  

The results of poetry generators are evaluated against Manurung’s (An evolutionary algorithm 

approach to poetry generation) three preconditions for successful AI-generated poems: (i) 

grammaticality, (ii) meaningfulness, and (iii) poeticness. The reasons for any poorness 

occurring are also looked into: the inability of computers to understand words as signs with a 

signified, a lack of general intelligence, process- (rather than output-) driven attempts, etc.  

Then, computer-like results (algorithmic processes) from a number of human poetic 

movements are also presented as a juxtaposition: Dada, stream of consciousness, OuLiPo, 

LangPo, Flarf, and blackout/erasure poetry. A remarkable similarity is found to surface 

between (i) human poets who are more concerned with experimentation than quality of results, 

and (ii) computer scientists who are process- rather than outcome-driven. This leads to a 

discussion of the characteristics of humanness, the possibility of granting future AI 

personhood, and the need to see our world in terms of a new and more refined ontology. 

http://journals.oregondigital.org/hsda/


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  2 

Introduction 

Attempts at poetry generation using Artificial Intelligence (AI) started as early as 1959, but it was already 

in the 1940s that humanities computing had given rise to Digital humanities. Digital humanties (DH) was (and 

is) a field that applies automation processes to texts, and this included attempts at poetry generation, without 

however ever making this its main goal. It was only after the turn of the millennium that computer science 

started seeing poetry generation as fertile ground for AI experimentation within the scope of Computational 

Creativity, a field that tackles several areas of art (such as prose, music, painting, song lyrics, photobashing, etc.) 

and science.  

If Digital Humanities’ attempts at poetry generation were restricted to rearranging or otherwise mixing 

human-written poems to produce something different, the question now with computer science was how to 

arrive at “intelligent” ways of generating poems rather than ruminating pieces of existing ones. To achieve this, 

complex and often knowledge-intensive systems that involved several computational and linguistic layers were 

employed. Creators of poetry generators today are less focused on the output; they are rather focused on the 

process of generation per se, a process that involves the teaching of AI to perform desired tasks. Thus, the 

potential contribution of humanities was rather disregarded and jostled altogether.  

Various attempts at poetry generation, either for fun or as academic projects, can be enlightening with 

regard to understanding the complex computational processes involved. But first, we will tackle the question 

of why poetry was chosen over other literary genres. 

Why poetry in the first place? 

Why did the creators of poetry generators choose poetry over another creative writing form, e.g. bonsai 

short stories (with a typical ceiling of around 600 words)? The answer lies in the characteristics of poetry relative 

to other literary genres. A short story, however small, must have a structure — characters, dialogues, plot, and 

so on. Poetry, on the other hand, incorporates a degree of arbitrariness since there are no strict or universal 

rules for what is acceptable or not. It also incorporates a degree of absurdity, as many a time it not only resists 

commonly acceptable meaning, but also reverses it. Despite being considered one of the most difficult literary 

genres to grasp due to its multitude of language levels (vocabulary, grammar, syntax, semantics, phonemics, 

metrics, figures of speech, etc.), poetry does not necessitate the use of all of them at the same time. In other 

words, it is not mandatory for all these elements to be present in a single poem, and this has certainly been 

noticed by computer scientists. Manurung (An evolutionary algorithm approach to poetry generation), for example, has 

noted that writing poetic text does not have to be extremely precise, as several rules of language that hold in 

the generation of natural language can be broken in poetry. 

The above characteristics theoretically make poetry more fertile ground for experimentation than any other 

literary genre. This is not only the case with AI; personal experience in literary criticism has shown that many 

new and ambitious human writers start off with poetry because they consider it more convenient and easier, 

and as a consequence, they usually begin by writing mediocre poems. Why should we expect any different from 

AI? 

Notwithstanding the above, there is also another reason to choose poetry as a field for experimentation: 

text- and reader-centered literary approaches have turned the headlights away from the author. The spotlight 

has shifted to the text first and the reader next — bringing about the “death of the author” (Barthes) and giving 

the reigns of determining the poem’s meaning to the other two. Among text-centered approaches and schools 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  3 

in literary theory, there are notably the following: Russian formalism (early 20th century), which looks for the 

meaning of a literary text in the mechanisms employed and its structural principles, i.e. the very qualities that 

make a text literary (Medvedev and Bakhtin); American New Criticism (1930s and on), which looks into the 

way that poetry interacts with the semantics of language (Richards; Eliot); and  Structuralism (1970s), which 

focuses on the conventions that allow literature to exist (Todorov). Under such approaches, the literary text is 

seen as an autonomous entity, the understanding of which requires nothing apart from its very words. 

Reader-centered approaches, on the other hand, include the following: Semiotics (late 19th century), which 

supports that the meaning of a literary text arises from the interaction of the sign, the interpreter, and the event, 

and can change over time (Culler); Psychoanalytic Criticism (1970s and on) which speaks of reading as a means 

of rebuilding the identity of the reader through a transactive relation that is established between him/her and 

the text (Holland);  Reception Aesthetics (Rezeptionsästhetik, 1960s), which maintains that the past experiences 

embedded in a text merge with the interests, cultural stock, and personal experiences of the readers as they read 

(Jauss);  and Reading Response Criticism (1970s), which talks of the ambiguity inherent in the text, creating 

gaps which the reader must bridge in order to arrive at an interpretation (Iser). 

The autonomy of the text, coupled with the reign of the reader as the assigner of meaning to it, caused a 

multitude of experimentations in literature to emerge. And much like computer scientists who are fascinated 

by the process, several poetic movements favored experimentation over a result that would fit mainstream 

poetry standards, challenging even the idea of such standards. This point will be addressed further, after a 

review of AI poetry generation history, methods, and processes.  

The baby steps of poetry generation 

First attempts to develop poetry using computers are more than 60 years old. The first poetry generation 

program is considered to be Theo Lutz’s “Stochastic Texts” in 1959. The program ambitioned to write poem-

like texts from scratch in the German language (Lutz). A sample of such a “poem” is the following: 

NICHT JEDER BLICK IST NAH UND KEIN DORF IST SPÄT. 

EIN SCHLOSS IST FREI UND JEDER BAUER IST FERN. 

JEDER FREMDE IST FERN. EIN TAG IST SPÄT. 

JEDES HAUS IST DUNKEL. EIN AUGE IST TIEF. 

 

NOT EVERY LOOK IS NEAR. NO VILLAGE IS LATE. 

A CASTLE IS FREE AND EVERY FARMER IS FAR. 

EVERY STRANGER IS FAR. A DAY IS LATE. 

EVERY HOUSE IS DARK. AN EYE IS DEEP. 

(Translation by Helen MacCormac, 2005) 

Same as with every other sample in the paper, this sample is seen through the lens of Manurung’s (An 

evolutionary algorithm approach to poetry generation) three preconditions for successful AI-generated poetry; 

Manurung said that AI-generated poems should include language that ensures the three elements of (i) 

grammaticality, (ii) meaningfulness, and (iii) poeticness. According to these preconditions, it is obvious that the 

results of the “Stochastic Texts” may be grammatical, but they are otherwise poor in terms of meaning and 

even poorer in terms of poeticness. Of course, one could argue that it was the first such attempt, after all. And 

perhaps it was remarkable to arrive even at such a result at the time.  



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  4 

Digital Humanities vs Computational Creativity 

Other early attempts at poetry generation mainly involved rearrangements or imaginative combinations of 

existing human-written poems. This was perhaps the result of the nature of the framework in which such poetry 

generation occurred: DH developed out of humanities computing (Henrickson), which has origins dating back 

to the 1940s and 50s vis-à-vis the pioneering work of scholars such as Roberto Busa and Josephine Miles 

(Stagnaro; Wimmer). 

Busa met Thomas J. Watson, the founder of IBM, in 1949 and convinced him to sponsor the Index 

Thomasticus project. The project involved a computer-generated concordance to Thomas Aquinas’s writings, 

and many people today still consider it to mark the onset of the field of computational linguistics. Josephine 

Miles, on the other hand, a poet and professor of English at Berkeley, became project director of an abandoned 

index-card-based concordance to the poetical works of John Dryden in the early 1950s. Partnering with the 

Electrical Engineering department at Berkeley and using their IBM tabulation machine, Miles used automatic 

methods to complete the concordance. This was eventually published in 1957, around the time that Busa 

circulated early drafts of his own concordance to Aquinas’s writings (Wimmer). 

Following these pioneers, other scholars from the humanities began using mainframe computers to 

automate tasks like word-searching, sorting, and counting, which was much faster than processing information 

from texts manually. All of these tasks involved DH. DH also touched on attempts at poetry generation, 

although never as holistically and persistently as AI is doing today. A notable example of DH poetry generation 

was the French “Atelier of Literature Assisted by Maths and Computers” (ALAMO). ALAMO sprang in 1981 

out of a poetic movement of the time, the OuLiPo, which was an acronym of Ouvroir de Littérature Potentielle, 

i.e. Workshop for Potential Literature (Oulipo; Berkman). An illustrious attempt of the ALAMO includes the 

rimbaudelaires, in which the altered structure of a sonnet by Rimbaud, acquired by cutting out nouns, verbs and 

adjectives, was filled in with words from Baudelaire’s poems (Laws). Rimbaudelaires are a strange mixture, 

producing lines that echo 19th-century French poetry and rendering rather solid results in terms of both 

coherence and meaning. Seen against Manurung’s (An evolutionary algorithm approach to poetry generation) 

preconditions (once more: grammaticality, meaningfulness, and poeticness), we may say that rimbaudelaires are 

good in terms of grammaticality, but they lack coherence (thus meaningfulness) in the succession of ideas, even 

though there is an allure of poeticness. 

A good example of rimbaudelaires is the following (Laws): 

Le Rêveur du bonheur 

C’est un lac de poitrine où passe une gamine 

Embrassant librement aux anges des sommeils 

D’argent; où le plaisir de la caresse fine 

Fuit: c’est un poudreux bonheur qui se rit de soleil  

Which roughly translates into: 

The Dreamer of happiness 

Is a bosom lake where a youngster passes 

Kissing freely the angels of sleeps 

Silver; where the pleasure of the fine caress 

Flees: it's a powdery happiness that laughs at the sun 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  5 

After the turn of the millennium, computer science started seeing the field of poetry generation as fertile ground 

for AI experimentation. A sound result was suddenly secondary in importance. The process was what mattered 

the most; how could AI arrive at “intelligent” ways of generating satisfactory poems? Trying to answer these 

questions, AI experts put to work language generation systems. These were often knowledge-intensive and 

involving several layers of language, such as phonetics, diction, syntax and semantics (Gonçalo Oliveira, 

Multilanguage extension and evaluation of a poetry generator), and several computational methods. The sad thing was 

that the role of humanities in the process was completely upended. This is probably due to “the cultural and 

historical divisions between artists, humanists, engineers, and scientists” (Roque). It is sad, however, because 

humanists might tangibly help computer scientists keep a keener eye on the output of poetry generation as well 

as the input. 

Online poetry generators for fun 

There are numerous sites that produce poetry online mainly for recreation as a game. Some of them are 

Poem Generator (Poem Generator),1 Language is a Virus (“Poetry”),2 Fantasy Name Generators (for haiku) 

(“Haiku Generator”),3 Acrostic Poem Generator (for acrostic poems) (Acrostic Poem Generator),4 The Genuine 

Haiku Generator (“The Genuine Haiku Generator”),5 Poem Portraits (Delvin),6 and Verse by Verse (Verse by 

Verse).7 Other projects, such as Poem Creator, which mixes up Shakespearean poems with Hip Hop songs, are 

available only as an iPad app. 

Let’s see an example of Google’s AI poetry:8 

this was the only way. 

it was the only way. 

 
1  This poem generator can generate many types of poems: free verse, haiku, rhyming couplets, sonnets, limericks etc.  

Poems are based on keywords that the user provides. See https://www.poem-generator.org.uk/ All links in this 
paper were last accessed on 22.7.2021. 

2  This site avails of a number of different poetry engines that include haiku and sonnets. It also avails of an interactive  
haiku generator that uses words which the user provides and are incorporated into templates created from famous 
haiku poems. See http://www.languageisavirus.com/poetry.php 

3  A random haiku generator that respects the standard (as perceived in the West) 5-7-5 syllable structure of the haiku.  
It also contains a disclaimer: “due to the randomness of a generator not all results will make perfect sense from a 
grammatical and/or strictly linguistic point of view”. See https://www.fantasynamegenerators.com/haiku-
generator.php 

4  Generator of acrostic poems (i.e. poems that spell out a word down the left side of the poem, beginning a new line  
with each letter of the word). The generator uses the name that a user enters to generate the poem. See 
http://www.acrosticpoem.org/ 

5  Random pseudo haiku generator. Using a Java Script application, this page automatically generates about 7.62e+39  
different haiku poems which, however, are admitted to be loaded with “too many incompatible images”. As for the 
generation mechanism, on the “What is this?” pad on the site, it is stated that “There are no predefined sentence 
structures; no pre-packaged lines. The words themselves decide which type of word will follow, as randomly as 
possible, while still maintaining some semblance of adherence to grammar and readability.” See 
http://www.everypoet.com/haiku/ 

6  Google’s AI poetry generation project, undertaken with researchers at Stanford and the University of Massachusetts  
is “an experiment at the boundaries of AI and human collaboration”. Users are asked to donate a word and become 
part of an ever-evolving collective poem. If they also allow use of their camera, they will receive a photo portrait of 
theirs accompanied by a two-verse poem. See https://artsexperiments.withgoogle.com/poemportraits 

7  Google’s AI poems inspired by famous American poets of the 19th century. The user selects a poet from a list and  
then designs the desired poem structure. Then, s/he writes the first line and chooses the rest from a list of 
suggestions. See https://sites.research.google/versebyverse/ 

8  More at: https://www.writerswrite.com/google-ai-writes-poetry-51920161 

https://www.poem-generator.org.uk/
http://www.languageisavirus.com/poetry.php
https://www.fantasynamegenerators.com/haiku-generator.php
https://www.fantasynamegenerators.com/haiku-generator.php
http://www.acrosticpoem.org/
http://www.everypoet.com/haiku/
https://artsexperiments.withgoogle.com/poemportraits
https://sites.research.google/versebyverse/
https://www.writerswrite.com/google-ai-writes-poetry-51920161


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  6 

it was her turn to blink. 

it was hard to tell. 

it was time to move on. 

he had to do it again. 

they all looked at each other. 

they all turned to look back. 

they both turned to face him. 

they both turned and walked away. 

Judged against Manurung’s preconditions (An evolutionary algorithm approach to poetry generation), we see that 

Google is doing well with grammaticality, but not so much so with meaningfulness (there is little coherence 

and a significant amount of repetition that obscures meaning), let alone poeticness. Like this one, most results 

from such online poetry generators are generally clumsy. Additionally, they may require a generous donation 

of words or even chunks of speech from the user, according to what they would prefer to see in the generated 

poem. In the above example, “it was the only way” was the donated phrase. 

Poetry generators as academic projects 

Besides the fun of the absurdity (or the occasional random success) of online poetry generators like these 

previous ones, poetry generation has also been undertaken as part of meticulous academic projects. Such 

projects used a variety of methods, much like poetry generators for fun have used, only in the academic 

framework, they have become more transparent. Among early academic projects were PoeTryMe (Gonçalo 

Oliveira, “Po”TryMe"),9  Poevolve (Levy),10  Tra-La-Lyrics 2.0 (Gonçalo Oliveira, “Tra-la-Lyrics 2.0”; Gonçalo 

Oliveira, et al., “Text based on Rhythm”), and ASPERA (Gervás, “Generating poetry from a prose text”). A 

more up-to-date joint project would be Deep-Spear,11 which produces lines like the following ones: 

that is the world art we a lord of god 

give in him, evermore to teach my prayers 

and only to forget her for its sake 

it is not love, for all thy flesh and peace. 

Against Manurung’s preconditions (An evolutionary algorithm approach to poetry generation), it is evident that with 

Deep-Spear, grammaticality is achieved (albeit with a Middle English accent), but meaningfulness is not (certain 

phrases are simply nonsensical) and neither is poeticness. Thus, the output is again rather poor. 

Lamb et al. (“A Taxonomy of Generative Poetry Techniques”) have devised a taxonomy which separates 

between “mere” poetry generation and result enhancement. In mere generation, a computer produces poems 

based on existing human-written ones and an algorithm that prescribes how the former might be used. As 

stated in the paper, “the results of mere generation can appear nonsensical; this is not always a bad thing from 

an artistic perspective” (Lamb et al. “A Taxonomy of Generative Poetry Techniques”). This is an interesting 

point that is worth some pondering. Inasmuch as we are ready to accept that a human can produce nonsensical 

 
9  Developed by the Center of Informatics and Systems of the University of Coimbra, Portugal. The generator comes  

with a diagram describing its operation. See: https://poetryme.dei.uc.pt/ 
10  See a detailed account by its creator at https://github.com/rplevy/poevolve 
11  Deep-Speare is a joint neural model of poetic language, meter and rhyme. See Lau et al.  

https://aclanthology.org/P18-1181/ and https://aclanthology.org/P18-1181.pdf 

https://poetryme.dei.uc.pt/
https://github.com/rplevy/poevolve
https://aclanthology.org/P18-1181.pdf


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  7 

poetry as a result of experimentation (because we know that they are capable of also producing poetry that 

makes perfect sense at the same time), we are less willing to accept that computers produce nonsensical poetry 

because it is their aim to do so. The primary reason for any nonsense is that AI uses words without 

understanding their meanings. AI may be able to tell that in the phrase “She has an apple”, “she” is a pronoun, 

“has” is a verb, and “apple” is a noun, or that “she” is the subject, “has” is the verb and “apple” is the object. 

But it cannot understand what this phrase is actually referring to in the real world nor draw a mental image, as 

the human brain does, of a woman holding an apple.  Words for AI are just data sequences, referring to no real 

information (Harel and Rumpe). In Saussurian terms, we would say that words for AI are one-dimensional 

signs; they involve just the signifier, without having a clue on what the signified is. 

Bostrom and Yudkowsky agree to that when they say that what AI lacks is generality. AI is trained to 

perform a single task. In this way, it resembles other-than-human biological life; in the example that the authors 

use, beavers can build dams and bees can build hives, but beavers cannot build hives and bees cannot build 

dams. A human watching beavers and bees, however, can build both, because human intelligence is applicable 

more generally, reflecting this generality. This has also been noted by the European Parliament Research 

Service, which has included in its report that a long-term goal of AI is the so-called Artificial General 

Intelligence (AGI).  

Methods of mere poetry generation 

Among the methods of mere generation that Lamb et al. (“A Taxonomy of Generative Poetry 

Techniques”) mention are the following: 

(a)     Templates. Template generation, also called slot-filling, has seen a wide application, as was the case 

with the rimbaudelaires we saw above. Templates can be extracted automatically from text, if lists of words or 

phrases pertaining to different categories (e.g. nouns or verbs) are drafted. Then, line templates with slots into 

which words from given lists can be inserted are created. Finally, words from the appropriate list are selected 

randomly to fill in the slots. 

Otherwise, templates can be handcrafted. Colton et al., for example, have used 

very short phrases (similes) mined from the internet, alongside the phrases of professional writers, 

namely journalists writing for the British Guardian newspaper. The former phrases fit into the long-

standing tradition of using the words of the common man in poetry, and the latter reflects ‘the desire 

to increase quality while not appropriating text intended for poems.’ (97) 

Theirs was a template-based approach which produced not only poetry but also brief stories and comments 

that accompanied each poem. This is because they identified the problem of computers having not been taken 

seriously as creative entities to be one of a lack of appropriate context. Thus, they saw that perhaps a story 

generator, able to produce “static framing information” and (in the future) to “reply with a story to any question 

asked of it in a dialogue situation” (101), would compensate for the lack of “human quality” in generating 

poems meant —after all— to be read by humans. The following is an example of a poem they generated, with 

the story first and then the poem entitled “Blue overalls,” following: 

It was generally a good news day. I read a story in the Guardian culture section entitled: “South Africa's 

ANC celebrates centenary with moment in the sun”. It talked of south africans, interfaith prayers 

and monochrome photos. Apparently, “The heroic struggle against a racist regime was remembered: 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  8 

those thousands who sacrificed their lives in a quest for human rights and democracy that took more 

than eight decades” and “At midnight he watched with amusement as Zuma lit the centenary flame, 

at the second attempt, with some help from a man in blue overalls marked ʻExplosivesʼ”. I wanted 

to write something highly relevant to the original article. I wrote this poem. (101) 

Blue overalls  

the repetitive attention of some traditional african chants  

a heroic struggle, like the personality of a soldier  

 

an unbearable symbolic timing, like a scream  

blue overalls, each like a blueberry  

some presidential many selfless leaders  

 

oh! such influential presidents  

such great presidents  

blueberry-blue overalls  

 

lark-blue overalls  

a knight-heroic struggle (Ibidem) 

The story is better than the poem in terms of meaning, though the algorithm becomes transparent because 

of the irrelevance of the two quotes, which give an overall awkward impression. The poem is weaker because, 

judged against Manurung’s (An evolutionary algorithm approach to poetry generation) preconditions for sound AI-

produced poetry, it has little meaningfulness (one of the main issues with template-based poetry generation). 

As Lamb et al. (“A Taxonomy of Generative Poetry Techniques”) point out in their discussion of templates, 

this poem too suffers from useless repetitiveness, but there is also little coherence in the example. And even 

though grammaticality is ensured, poeticness is present only in the form of the poem, i.e. the stanzas that create 

a poetic impression. Therefore, the problem is not one of context, as Colton et al. claim; the problem is one of 

meaning. Meaning, as we have seen from the examples so far, remains elusive for AI. As to the specific method, 

it should be added that the creativity of template-based generation is “open to discussion” (Gervás “Exploring 

Quantitative Evaluation of the Creativity of Automatic Poets”), as the input is pre-defined and the choices 

involved are narrow. 

(b)     Markov chains. A Markov chain is a statistical model applied to data in a series, calculating the 

probability distribution for each entry. The N entries used to make the probability prediction are each referred 

to as an n-gram. For poetry generation, entries would be words. Entries could also be letters but that would 

produce a multitude of non-words and neologisms, an effect which could be intentional but is not successful 

in terms of outcome. An example of characters used as entries comes from Roque; he introduces a poetry 

generator developed by R. W. “Bill” Gosper in the early 1970s at the Artificial Intelligence lab at the 

Massachusetts Institute of Technology, using character n-gram generation. Gosper approach was later called 

Dissociated Press. The following example cited by Rogue is a representative output of Dissociated Press on 

Lewis Carroll’s Alice in Wonderland.12  

 
12  A probability model can be generated from a specific work, as in this case, or from a broader corpus. 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  9 

     (…) 

Would guess, an’t 

grom one foot to thistle, 

to keep and reachinah’ll be she could not ever who had not atte-book 

hastily. Convers began to trings: into thing on wast the door, and 

the. 

Involving an automatic shuffling of words, as it does, Markov chain poetry can be said to resemble magnetic 

poetry13 or even Dadaist “cut-ups.”14 A project that involves Markov chains is EMILY (Shihadeh and 

Ackerman). EMILY is a machine that aims to create original poems in the style of Emily Dickinson, and it was 

trained on publicly available Emily Dickinson poetry from the Gutenberg project. With the help of Markov 

chains, poems in the style of Emily Dickinson were attempted, relying on a dictionary that maps words found 

in Dickinson’s poems to the words that proceed them to capture the relationships between them. An example 

of a Dickinson-like poem produced by EMILY would be the following:  

Poem 1 

Some shook their yellow gown 

And certainly her eye, they  

Leap upon the rose smiling  

To die. The orchards Eternity! 

Applying again Manurung’s (An evolutionary algorithm approach to poetry generation) preconditions for sound AI-

produced poetry, we would say that the result is rather poor. Even though grammaticality is achieved (save for 

the last line), meaningfulness is not, and poeticness exists only as an impression that borders ridicule. A weak 

point of Markov chains is that even though they preserve many features of the input text, they cannot replicate 

grammar. 

(c)     Found poetry. This method involves skipping the generation process altogether and using a 

computer to harvest text written by humans. Found poetry preserves entire human-written sentences that are 

presented outside their original context. A splendid example of found poetry are the haikus generated by 

Hrešková and Machová. 

Hrešková and Machová have produced haikus using two different approaches: an evolutionary algorithm 

and poem models. In the evolutionary algorithm approach,15 a corpus of haikus was built from two haiku 

portals containing human-written haikus. Cross-overs, i.e. exchanges, were attempted from two or three 

existing haikus randomly. This corpus was then evolved by human judges rating every attempt. Examples of 

such haikus are as follows: 

 
13  Magnetic poetry, devised by Dave Kapell, is based on moving around on a metal board words that are written on  

magnets. See the official website of this poetic movement https://magneticpoetry.com/ 
14  At a surrealist rally in the 1920s Dada poet Tristan Tzara proposed to create a poem on the spot by pulling words  

out of a hat. Almost 40 years later, in the summer of 1959, painter and writer Brion Gysin cut newspaper articles 
into sections and rearranged the sections at random. “Minutes To Go” resulted from this initial cut up experiment. 
“Minutes To Go” contains unedited unchanged cut ups emerging as quite coherent and meaningful prose. See 
Burroughs, “The Cut Up Method.” 

15  Evolutionary algorithms are just one of the various methods used in poetry generation. For other approaches, see  
Lamp et al., “A Taxonomy of Generative Poetry Techniques”; Gonçalo Oliveira, “PoeTryMe: a versatile platform 
for poetry generation”; Gervás, “Exploring Quantitative Evaluation of the Creativity of Automatic Poets.”) 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  10 

spring wind 

an old lady tries again 

wet with rain  

 

aging beauty 

spring cascades into valleys 

as apple blossoms 

The haikus achieved are robust and convey a true sense of “haikuness,” i.e. poeticness according to the 

norms of the specific genre. Both grammaticality and meaningfulness are also achieved. This is not to say that 

less meaningful haikus or haikus with confusing content were not created but that these were rather the 

minority. As enthusiastic as one can be with such a great result, the fact that whole lines of existing haikus were 

recombined bodes poor for originality or creativity. 

But the success of the result makes us understand that, considering mere generation methods, the quality 

of input is paramount. Many generators take as input content from newspapers (such as Colton et al.), blogs 

(e.g. Wong and Chun), psychological tests (Netzer et al.), even the Twitter (e.g. the TwitSong, Lamb et al., “Can 

human assistance improve a computational poet?”). This does allow “using the words of the common man in 

poetry” (Colton et al. 2012), but it is far from a poetic result. The main reason is that not all words of common 

man have poetic value. By contrast, when the input is highly relevant, as with the haikus generated by Hrešková 

and Machová, the result is more in line with a sound poetic production, even though it is poorer in terms of 

creativity. 

What should be preferred, then? An intensely human-driven poetic result, which is less creative in terms 

of the auxiliary role of AI, or a bad poetic result arrived at by AI alone? It could be argued that poetry generators 

should be about producing a sound poetic output (otherwise, why poetry generators?). But the truth is that it 

should not be poetic at all costs. The real issue of AI poetry generation is that it should be intelligent; in other 

words, the computer should be able to actually do it on its own. Thus, the process by which a result is achieved 

is as important as the result itself. Of course, the result is ultimately the measure by which the success of the 

attempt is rated. And the result often is, as we have seen, poor, especially so in terms of meaningfulness. 

Enhancement of poetry generation results 

Moving away from mere generation, Lamb et al. (“A Taxonomy of Generative Poetry Techniques") next 

tackle the issue of enhancement with regard to poetic results. Enhancement may be conducted either by a 

person or by the computer itself. 

Human enhancement 

As is obvious, a human can enhance the output of poetry generators, editing the poems themselves. 

Although this might ultimately render the generator useless in the first place, it is undeniably an established 

practice. It works marvelously for those who see computational text generation as nothing more than a good 

starting point for writing poetry, equal to the provision of raw material.  

 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  11 

The practice of human enhancement is evident in Gnoetry (Gnoetry Daily), an application for interactive 

text generation in which some decisions are made by the computer and others by a human user. Generated 

poetry is based on n-grams, but it is the user that provides the corpus. Thus, a dialogue is established between 

the human and the computer (Lamb et al. “A Taxonomy of Generative Poetry Techniques”). An example from 

Gnoetry (Gnoetry Daily) is found below:16 

 

The day stirs the bare limbs above me, 

dry breath from the frozen fields.  

                                                              Inside, 

charred, still smoldering, I sit patiently 

  

with my despair, having done too little, 

for too long, to fan these flames, having been 

  

content enough, secure enough in my same 

  

fearful ways, narrow, shut in, obscure 

to myself. 

Computer enhancement 

Poetry generation results can also be enhanced by the computer itself. As pointed out by Lamb et al. (“A 

Taxonomy of Generative Poetry Techniques”), “this set of methods comes not from the humanities but from 

scientists in the discipline of computational creativity” (198) and involves two main purposes. One is 

“optimization of the system’s output on some metric,” and the other is connection to underlying knowledge 

about the real world. 

Beginning with the latter, Lamb et al. (“A Taxonomy of Generative Poetry Techniques”) acknowledge that 

the nonsensical results of mere poetry generation are the “result of the computer’s lack of real-world 

experience” (198). It is the problem with the phrase “She has an apple,” as described above. The computer may 

be able to say which word is what kind of thing in terms of grammar or its place in a syntactic tree, but it cannot 

understand what this phrase is really about. 

This is a serious issue and the real reason behind the production of nonsense, yet this can be solved to 

some degree by the provision of knowledge bases. In poetry generation, a knowledge base programmed into 

the generation process may guide its output. This is done in much the same way as e.g. in Gnoetry (Gnoetry 

Daily), where human input inside the generation process optimizes results. In particular, a knowledge base built 

on the calculation of the co-occurence of different words in a source text can give the computer a sense of 

which words go better together or are more relevant to a topic. Knowledge bases may also be extracted from 

dictionaries and cover relations between words, such as junction, composition, opposition, causation, etc. 

(Gonçalo Oliveira, “PoeTryMe: a versatile platform for poetry generation”). Netzer et al. have produced quite 

good haikus from a list of word associations that are combined on the basis of syntactic templates: 

 
16  More can be retrieved at https://gnoetrydaily.wordpress.com/ 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  12 

early dew 

the water contains 

teaspoons of honey 

 

spring bloom  

showing  

the sun’s pyre 

Regarding optimization, a computer may be fed with some formal definition of the desired characteristics 

in a poem and can thus test different possibilities in order to choose the ones that best fit such characteristics. 

These are baby steps to critical thinking, but it is a start. Lamb et al. (“Can human assistance improve a 

computational poet?”) list several optimization methods, some of which are as follows: stochastic hill-climbing 

search, generate-and-test, case-based reasoning and recurrent neural networks. Many such methods concentrate 

on issues of poetic form (e.g. rhyme and meter), with fewer on meaning, style, or emotions. TwitSong (Lamb 

et al. (“Can human assistance improve a computational poet?”) browses tweets to choose those that are relevant 

to a topic and produce sentiment. The process generates poems that are composed of tweets which, although 

written by different people at different occasions, make up something reasonably meaningful, even though its 

poetic value could be strongly challenged: 

          Hey Nashville… 2014 is pretty awesome! 

Happy 2014 friends! Be safe out there!! 

           Had a great New Year’s Eve at Magic Kingdom 

           We started off 2014 with a prayer. 

Lamb et al. (“Can human assistance improve a computational poet?”) conclude that poetry generation is 

still at an infancy stage and that there is a lot to be expected. Hopefully, that will be the case. Why hopefully? 

Because, it could lead to a new style or a new content of poetry, which would be simply refreshing.  

Poetry generation in terms of target features 

Lamb et al. (“A Taxonomy of Generative Poetry Techniques”) are not the only ones with a taxonomy of 

poetry generation systems; other scholars have too provided their own taxonomies. For example, Gervás 

(“Exploring Quantitative Evaluation of the Creativity of Automatic Poets”) has presented four types of AI 

techniques employed in the generation of poetry and Gonçalo Oliveira (“A Survey on Intelligent Poetry 

Generation: Languages, Features, Techniques, Reutilisation and Evaluation”) has tackled the point from a 

different perspective, allowing a closer glimpse at the workings of generators. More particularly, Gonçalo 

Oliveira makes a useful distinction between methods used to address the form features considered by poetry 

generators and methods used to address content features, both of which the following section will explore. 

Form features 

Form features play a prominent role in poetry generation, as they allow recognition at first glance that the 

output is, in fact, a poem. Meter and rhyme are such form features that can be quite competently handled by 

computer programs. Meter involves the number of syllables that each line of the poem must have, taking into 

consideration stress patterns. Haikus, for example, have three lines that include 5-7-5 syllables, respectively (as 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  13 

is perceived in the West, at least), and limericks typically have 5 lines. Moreover, sometimes stress patterns are 

also important. 

Rhyme is the repetition of the same sound in different lines, according to the appropriate scheme, e.g. 

AAAA (monorhyme), AABBCC (coupled rhyme), ABAB (alternate rhyme), ABBA (enclosed rhyme), AABBA 

(limerick), or ABA BCB CDC etc. (terza rima, or chain rhyme). Some systems deal with other kinds of rhyming 

apart from end-rhyme, e.g. assonance or alliteration. These involve the repetition of the same vowel or 

consonant sound, respectively, in some parts of the poem or throughout it.  

According to Gonçalo Oliveira, for “less phonetic languages such as Portuguese or Spanish […] it is often 

enough to design a set of orthography-based rules to handle meter and rhyme” (Gonçalo Oliveira et al., “A 

Survey on Intelligent Poetry Generation”; Gervás, “An expert system for the composition of formal Spanish 

poetry”). For English, poetry generators typically resort to pronunciation dictionaries, often online ones (see 

e.g. Manurung, “A chart generator for rhythm patterned text”; Tobing and Manurung , “A chart generation 

system for topical metrical poetry”). 

Content features 

Even if form features are more easily tackled with AI, the issue of content remains paramount. No text can 

be considered a poem if it just follows a recognizable poetry form. It must contain language that, according to 

Manurung (An evolutionary algorithm approach to poetry generation), ensures the central three elements we have already 

discussed: (i) grammaticality, (ii) meaningfulness and (iii) poeticness. 

Out of these three, grammaticality seems to be the most easily achievable one; after all, natural language 

generation does involve syntactic rules, and lexical-syntactic patterns can also be achieved. Meaningfulness, on 

the other hand, is more elusive. Meaning is only loosely achieved in AI by using, for instance, words that belong 

to the same semantic domain (e.g. fruit-apple-orange-mango and other hypernymic-hyponymic relations, or 

synonyms/antonyms etc.) or by creating semantic clusters of words that are connected to each other in various 

ways (for instance, successful combinations of adjective and noun, verb and noun). Such techniques may allow 

for the generation of other natural language texts, such as weather forecasts. For a weather forecast to be 

generated, there must be a computer that is able to recognize weather phenomena, e.g. storms, precipitation, 

hurricanes, and their movements. Putting all these into a forecast would then require the computer to be able 

to name the weather phenomena, select interesting pieces of information while rejecting other not-so-

interesting ones, and decide on a good textual structure, among other things (Wright and Purver). 

In poetry generation, meaning is attempted by exploiting a semantics model, a semantic knowledge base, 

or a statistical model of distributional semantics (Gonçalo Oliveira “Multilanguage extension and evaluation”).  

Semantics models are based on labelled relations between words and reflect a more theoretical view of linguistic 

knowledge. Knowledge bases we have already discussed above in “Computer enhancement.” Distributional 

models of semantics consider how language is actually used, e.g. in a collection of texts, and assume that words 

occurring in similar contexts have similar meanings. In other systems, text is generated according to a grammar 

that handles syntax (Manurung An evolutionary algorithm approach to poetry generation) or a grammar that is tightly 

related to semantics (Gonçalo Oliveira, “PoeTryMe: a versatile platform for poetry generation”).  

Mere text generation, however, simply does not suffice. The poem has to be meaningful, i.e., it must be 

about something, otherwise it must at least be able to respond to various stimuli ((Gonçalo Oliveira, “A Survey 

on Intelligent Poetry Generation”). Such stimuli could be any of the following: a list of semantic predicates, e.g. 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  14 

love (John, Mary) (Manurung An evolutionary algorithm approach to poetry generation); one or more keywords that will 

set a semantic domain (e.g. Netzer et al.; Gonçalo Oliveira, “PoeTryMe: a versatile platform for poetry 

generation”); a text which can be a single sentence with a message (Gervás “An expert system for the 

composition of formal Spanish poetry”) such as a chunk taken from a newspaper (Colton et al.).  

As for poeticness, figurative language is often implicitly present as a consequence of using human-produced 

poetry as input. We have seen the apt example of Hrešková and Machová’s haikus. There is, however, ongoing 

research in the field of figure-of-speech generation, e.g. regarding the generation of similes (as-as or like) mined 

by using Google n-grams (Veale) or the dynamic generation of similes from scratch (Chakrabarty, Muresan & 

Peng).17 Such similes are then used for the generation of other similes or metaphors or yet other figures of 

speech. But in real poetry, using figures of speech alone does not suffice; a supporting poetic context should 

also be in place. 

Human or Not 

Poetry generators are not the only “poets” that produce strange results. Like the computer scientists 

discussed earlier who were more interested in process than product, several modern poetic movements have 

favored experimentation over a result that would fit mainstream poetry standards, challenging even the idea of 

them. This certainly makes poetry generation look like another link in the long chain of poetic tradition, as will 

become clearer at the end of this section. Some of the most striking such movements, involving algorithmic or 

algorithmic-like processes, are mentioned below. 

Dada 

As early as in 1916, Dada made its appearance as a poetic movement of the European avant-garde. Coming 

into being around of World War I, its founders tried to “embody a simultaneously playful and nihilistic spirit 

alive among European visual artists and writers” (“Dada”).  They attempted to salvage a sense of freedom from 

the instability that characterized European society in the aftermath of the war, embracing both “everything and 

nothing” in their desire to “sweep, sweep clean,” as Tristan Tzara wrote in the Dada Manifesto in 1918. These 

Dada poems were made of cut-ups, combined in the form of collage to produce a syntactic result that 

encouraged readers to question what they were reading and make up their meanings for the poems’ words. 

The following is a representative Dada poem entitled “Assessment” from Three Poems by Tristan Tzara: 
  

 
17  The achievement of Chakrabarty et al. makes us realize that the real benefit lies in seeing through human brain  

processes and trying to reproduce them, as if reinventing them. This would necessarily also make AI another way of 
understanding human intelligence better. 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  15 

 

Considering that the origins of this poem are from different sources, as is evident by the different styles of 

the cut-ups, one could argue that — but for the stylistics — a poem generator that uses human-written texts as 

input seems to walk on a very similar path to Dada. 

Stream of consciousness 

Stream of consciousness is another notable example of literature written without much concern for the 

outcome. The term “stream of consciousness” was coined by the psychologist William James in The Principles 

of Psychology (1890). This style intended to render “the flow of impressions — visual, auditory, physical, 

associative, and subliminal — that impinge on the consciousness of an individual and form part of his awareness 

along with the trend of his rational thoughts.”18 This technique was adopted by many poets, one of which is 

notably Gertrude Stein. Here follow two poems of hers: 

Red Faces 

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.19 

 

America  

Once in English they said America. Was it English to them. 

 
18  https://www.britannica.com/art/stream-of-consciousness  
19  https://allpoetry.com/Red-Faces  

https://www.britannica.com/art/stream-of-consciousness
https://allpoetry.com/Red-Faces


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  16 

Once they said Belgian. 

We like a fog. 

Do you for weather. 

Are we brave. 

Are we true. 

Have we the national colour. 

Can we stand ditches. 

Can we mean well. 

Do we talk together.20 

Both poems are asyntactic to some extent and contain a flow of ideas that is not easy to follow. They 

resemble very much the current output of poetry generators. 

OuLiPo 

OuLiPo, another poetic movement mentioned in “Digital Humanities versus Computational Creativity,” 

emphasized systematic, self-restricting processes of making texts that resemble algorithms. For example, the 

technique known as n+7 replaced every noun in an existing text with the noun that followed seven entries after 

it in the dictionary. By applying the n+7 rule to Wallace Stevens’s poem “ (Stephens), for example, one gets a 

completely new poem, “The Soap Mandible” (“A Brief Guide to OULIPO”): 

The Soap Mandible 

One must have a miniature of wisdom 

To regard the fruit and the boulders 

     Of the pinions crusted with soap; 

 

And have been colic a long time 

To behold the junkyards shagged with Idaho, 

The spun-yarn rough in the distant gloom 

 

Of January surgery; and not to think 

Of any mishap in the south of the winter, 

In the south of a few lectures, 

 

Which is the south of the language 

Full of the same winter 

That is blowing in the same bare plague 

 

For the lithographer, who listens in the soap, 

And, now himself, beholds 

Now that is not thermal and the now that is. 

 
20  https://allpoetry.com/poem/8529995-America-by-Gertrude-Stein  

https://allpoetry.com/poem/8529995-America-by-Gertrude-Stein


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  17 

Another OuLiPo exercise uses the “snowball” technique, where the first line is one letter long, the second 

line has two letters, and so on. In the 15-line snowball poem that follows by Harry Mathews, each line is n+1 

character longer than the previous: 

I 

am 

the 

text 

which 

begins 

sparely, 

assuming 

magnitude 

constantly, 

perceptibly 

proportional, 

incorporating 

unquestionable 

incrementations 

As becomes evident from the above samples, Oulipians were more interested in experimentation than in the 

outcome, as is often the case with art. Why should AI generated poetry be any different? 

Language Poetry 

Language Poetry, or LangPo, is yet another example of poetry that could have been written by a poetry 

generator. Among the key beliefs of LangPo lies the idea that language is what dictates meaning. LangPo poets 

also intend to make the reader actively involved in the text: “by breaking up poetic language, the poet[s are]  

requiring the reader to find a new way to approach the text” (“A Brief Guide to Language Poetry”). An example 

of a LangPo poem is the following, an excerpt from “Letter 7” written by Michael Palmer (Palmer): 

Letter 7 (excerpt) 

But the buried walls and our mouths of fragments, 

no us but the snow staring at us . . . 

 

And you Mr. Ground-of_what, Mr. Text, Mr. Is-Was, 

can you calculate the ratio between wire and window, 

 

between tone and row, copula and carnival 

and can you reassemble light from the future-past 

 

in its parabolic nest 

or recite an entire winter’s words, 

 

its liberties and psuedo-elegies, 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  18 

the shell of a street-car in mid-turn 

 

or scattered fires in the great hall 

I would say not-I here I’d say The Book of Knots 

 

I’d say undertows and currents and waterspouts, 

streaks of phosphorus and rivervine winds 

 

Dear Z, I’d say it’s time, it’s nearly time, it’s almost, it’s 

       just about, it’s long 

past time now time now for the vex- for the vox- for the 

       voices of shadows, 

 

time for the prism letters, trinkets and shrouds, 

for a whirl in gauzy scarves around the wrecked piazza 

Flarf 

A 21st century poetry movement reflecting similar themes is Flarf, which thrives on language from the web 

to arrive at odd juxtapositions and grammatical inaccuracies. Flarf also celebrates deliberately bad or “incorrect” 

poetry by forcing clichés, swear words, onomatopoeia, and other linguistic aberrations into poetic shape 

(“Flarf”). Flarf poets mine such odd terms from Google to arrive at often hilarious and sometimes disturbing 

poems, plays, and other texts (a technique called “Google sculpting”). The extensive use of internet resources 

is much akin to most poetry generators. An example of Flarf poetry is the following poem by Rodney Koeneke 

(Koeneke): 

bug’s psalm 

The bug’s psalm: don’t get crushed. 

Afterlives feel meaningless 

but spring will come, 

push out the nubs 

the kids braid into pallets. 

Take up your pallet 

from lawns noon’s hardly touched. 

The small think gods 

just loll on clouds. 

Bugs think gods just crush. 

The incoherence of this poem makes one think that it could actually have been written by a computer, but it is 

again an example of human-written poetry for the sake of experimentation. 

Blackout or Erasure Poetry 

Last but not least, blackout or erasure poetry can be said to resemble the workings of poetry generators, 

erasing as it does all the words in a given text save for those that it chooses to create a poem. It is therefore a 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  19 

process of narrowing down the words from a given text (newspaper articles are used very often; however, most 

any text will do) to a few ones according to what the poet sees fit (Glatch). An example of blackout poetry — 

rather bad, but helpful for the sake of the argument — would be the following: 

great success 

against time is 

You look 

on the outside 

like you feel 

 

Miller, E. Ce. “Blackout Poetry Is A Fascinating Art Form You Can Try At Home Right Now.” August 28, 
2017. https://www.bustle.com/p/what-is-blackout-poetry-these-fascinating-poems-are-created-from-existing-
art-78781 

All the foregone examples help to show that sometimes people, too, write poetry as if they were computer 

generators, not minding incoherence, elliptical meaning, or even nonsense being produced. This fact highlights 

the elusiveness of poetry and creativity; not only it cannot be solidly defined, but it also means different things 

to different people or groups of people over time. What’s more, AI attempts at poetry generation are not so 

different from similar human attempts. This means that, in a way, AI attempts are actually worth a place in the 

long tradition of poetry as it evolves. 

Human or Bot 

Having said all of the above, is it even possible to understand nowadays if a poem has been written by a 

human being or a computer? The answer is that sometimes, indeed, it is not. This is because some AI attempts 

have, in fact, proven to be of very good quality — and because human beings have written poems that can 

closely mimic computer attempts. So, if there are computers that write as computers, computers that write as 

human beings, human beings that write as human beings, and human beings that write as computers, “what on 

earth is going on”? This question was posed in a TedX speech by Oscar Schwartz, a writer, poet, and creator 

https://www.bustle.com/p/what-is-blackout-poetry-these-fascinating-poems-are-created-from-existing-art-78781
https://www.bustle.com/p/what-is-blackout-poetry-these-fascinating-poems-are-created-from-existing-art-78781


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  20 

of an online site on AI versus human-written poetry (Schwartz).21 Schwartz further talks about the capacity of 

being human and what can define it. Is it something with which we are born, or something that we are trained 

to become? And how do we define who — or what — is part of it? Generalizing therefore on the concept of 

AI and not restricting it to poetry, let us consider it, for a while, in relation to humanness.  

In 2017, six scientists wrote and signed a manifesto in which they also wonder: Is the capacity of being 

human something that can be lost or won? (Barzov, “Manifesto: Time for Humanness Learning”). Are those 

who torture, kill, humiliate human existence still considered human? Is a person born in a western country 

more human than an indigenous in, say, New Zealand? In the end, is human nature a social, thus changing, 

construct? And they end up saying: “It’s time for us to come out and start building a community of AI scientists, 

neuroscientists, philologists, game developers, historians of literature, movie producers, writers, entrepreneurs 

– all those who are united by the passion to discover the first principle of humanity and —furthermore— by 

the genuine desire to provide humans and machines with compelling and efficient tools of learning and 

relearning to be human” (Barzov, “Manifesto: Time for Humanness Learning”).  

But how can computers and AI “learn to be human” if they are not? The idea of granting “personhood” 

to AI has already become a point of scrutiny. To begin with, it has been the case for non-humans to be 

considered persons in certain countries. In the USA, corporations like the South Pacific Railway have been 

granted the rights of free speech and freedom of religion since 1881 (Winkler). In other countries, like New 

Zealand or Mexico, natural sites such as the Te Urewera Forest or Mount Taylor, respectively, have person-like 

rights (Colwell). On the other hand, there are law scholars today who have shown that anyone can assign legal 

personhood to a computer by having it run a limited liability corporation in the USA. In this way, it could own 

property, hire lawyers, sue, and enjoy protections provided under law as if it were a regular human being. And 

what about the robot Sophia to which Saudi Arabia granted citizenship, causing objections from Saudi women 

who noted that the robot was given more rights than them? (Yampolskiy)?  

Granting personhood to AI is an issue that has occupied a number of law scholars, also involving scholars 

from other fields like ethics (Garcia). Gunkel and Wales have recently looked at it from two different 

standpoints, one involving philosophy and law (Gunkel) and another involving theology and psychology 

(Wales). Gunkel (Gunkel and Wales) draw a useful division between natural and legal persons; they state that 

natural persons are those who possess the capability of speech, have consciousness, and are able to experience 

pleasure and pain, whereas legal persons are entities defined as such not due to their innate qualities but rather 

following a legal decision. Then, Gunkel cites “human beings of diminished intellectual capacity or in a 

persistent non-responsive state” (475) who are not full legal persons. This makes us understand that, although 

there are non-human entities that are considered legal persons as we saw above, there are also humans who, by 

contrast, are not considered to be such. One can also note at this point the suggestion of the European 

Parliament for AIs to be considered “electronic persons,” which was reflexively followed by an open letter 

signed by 250 scientists, engineers, and AI professionals opposing the suggestion and asserting that robots and 

AI are nothing more than tools (Open Letter). 

Wales (Gunkel and Wales), on the other hand, tackles the issue of the interior life of human beings, who 

engage in “voluntary self-gift” as they meet other human beings’ interiorities in a fusion of minds through 

 
21  That site was Bot or Not (botpoet.com). The user had to decide whether the poem that he saw on his screen was  

written by a computer or a human. It operated, in other words, on the basis of “Turing tests.” A Turing test, named 
after the computer scientist Alan Turing, is intended to show whether a machine can convince someone that it is 
human. Sadly, this site is no longer live. 



Humanist Studies & the Digital Age  Linardaki 

7.1 2022  21 

empathy and conscious understanding. Artificial neural networks, as he points out, are just biologically inspired 

simulations without any such characteristics or potential. AI, Wales argues, “is a sedimentary reflection of our 

own personhood. It is artificial in the original sense of that word – an artifact, a work of skill that we have 

brought forth by gazing into a computational pool of Narcissus” (480). In other words, even if we create 

computers with human form or capacities, they will remain mirrors of human beings, who — among other 

things — feature self-awareness and a sense of identity. These are qualities that something that a machine, 

which is programmed just to execute orders, cannot ever possess. Transhumanist thinkers have long pondered 

such issues (Bostrom, “A history of transhumanist thought”). Is it, then, ever imaginable for a mirror to reach 

or overcome that which it mirrors? This remains to be seen, literally. 

All these discussions and ideas are, of course, far-fetched when it comes to poetry generators that employ 

human-driven imaginative combinations of human-created examples to come to a result. Computational 

Creativity, at present, remains useful in helping humans with their own creative endeavors, even though its 

attempts at imitating human creativity have yielded interesting and at the same time ineffectual results. It does 

not possess imagination, however, nor inspiration of its own, and any talks about personhood in this case would 

only seem irrelevant (Barzov, “Human Intelligence (HI) is underrated”). The real value of Computational 

Creativity perhaps rests in what it can reveal about human creativity and cognition (Carleton University). 

But what about the future? According to Dormehl, “Computational Creativity has been ignored. Right 

now, either fondly or maliciously, it is being laughed at. Next it will start fighting our preconceptions.” It is true 

that when AI creates something, we have the tendency to credit its programmer alone. After all, machine 

creativity today typically involves humans making at least some of the decisions. However, the boundary 

between programmer and computer is becoming increasingly fluid and indistinguishable, as many of the 

computer methods of today produce results with production details that the programmer has no clue how to 

explain. The scholar Matthias corroborates this when he says that the programmer of a neural network, for 

instance, has little control past the point of coding. Soon, perhaps, the credit for any AI achievement will be 

split between humans and machines in much the same way as with films written by two scriptwriters, even if 

this will, in fact, mean that AI has to acquire legal personhood.    

Such a development will seem less and less bizarre as AI grows more and more capable of performing the 

duties assigned to it. After all, philosopher Immanuel Kant held that moral concern should extend equally to 

all rational beings, including ones that may appear from other planets. According to Boddington, Kant might 

also have added that it could apply to AI, if AI had had then the widespread presence it has today (13).  

It is also possible that the train of thoughts we currently follow is distracting us from perceiving the real 

opportunities and challenges which confront us in the age of AI (Gunkel and Wales). Maybe what we need is 

a more refined ontology that does not see the world in terms of black and white. And maybe it is Computational 

Creativity that will show us the way to it, which would make it truly invaluable. 

Works Cited 

“A Brief Guide to Language Poetry.” Poets.org, Academy of American Poets, 2004, 

https://poets.org/text/brief-guide-language-poetry. Accessed 22 July, 2021. 

“A Brief Guide to OULIPO.” Poets.org, Academy of American Poets, 2004, https://poets.org/text/brief-

guide-oulipo. Accessed 22 July, 2021. 

https://poets.org/text/brief-guide-language-poetry
https://poets.org/text/brief-guide-oulipo
https://poets.org/text/brief-guide-oulipo


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  22 

Acrostic Poem Generator, http://www.acrosticpoem.org/. Accessed 22 July, 2021. 

Barthes, Roland. “The Death of the Author.” Translated by Richard Howard, Aspen, no. 5-6, 1967, 

www.ubu.com/aspen/aspen5and6/threeEssays.html. Accessed 22 July, 2021. 

Barzov, Yuri. “Human Intelligence (HI) is underrated.” Medium, May 25, 2018, 

medium.com/@yuribarzov/humanness-is-underrated-d2b745124dd4. Accessed 22 July, 2021. 

Barzov, Yuri, et al. “Manifesto: Time for Humanness Learning.” Medium,  March 22, 2017, 

towardsdatascience.com/manifeto-time-for-humanness-learning-51051befa5f4. Accessed 22 July, 2021. 

Berkman, Natalie. “Digital Oulipo: Programming Potential Literature.” Digital Humanities Quarterly vol. 11, no. 

3, 2017. http://www.digitalhumanities.org/dhq/vol/11/3/000325/000325.html. Accessed 22 July, 2021. 

Boddington, Paula. Towards a Code of Ethics for Artificial Intelligence. Springer International Publishing, 2017. 

Print. 

Bostrom, Nick and Yudkowsky, Eliezer. “The ethics of Artificial Intelligence.” Cambridge Handbook of Artificial 

Intelligence. Eds. William Ramsey and Keith Frankish. Cambridge University Press, 2011. Print. 

Bostrom, Nick. “A history of transhumanist thought.” Journal of Evolution & Technology, vol. 14, no. 1, 2005. 

www.nickbostrom.com/papers/history.pdf. Accessed 22 July, 2021. 

Burroughs, William. “The Cut Up Method.” The Moderns: An Anthology of New Writing in America. Ed. Leroi 

Jones. New York: Corinth Books, 1963. https://www.writing.upenn.edu/~afilreis/88v/burroughs-

cutup.html. Accessed 22 July, 2021. 

Carleton University, DIGH500 cohort. Digital Humanities: A Primer, chapter 10, 2018-2019, 

carletonu.pressbooks.pub/digh5000/. Accessed 22 July, 2021. 

Chakrabarty, Tuhin, et al. “Generating similes effortlessly like a Pro: A Style Transfer Approach for Simile 

Generation.” Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 

2020, Online, November 16-20, 2020. Eds. Bonnie Webber, et al.  Association for Computational 

Linguistics, 2020. 6455-6469. Accessed 22 July 2021. 

Colton, Simon, et al. “Full-FACE poetry generation.” Proceedings of 3rd International Conference on Computational 

Creativity (ICCC 2012), Dublin, Ireland, University College Dublin, 2012. 95-102. 

https://www.computationalcreativity.net/proceedings/ICCC-2012-Proceedings.pdf. Accessed 22 July 

2021. 

Colwell, Chip. “What If Nature, like Corporations, Had the Rights and Protections of a Person?” The 

Conversation, 8 Oct. 2021, https://theconversation.com/what-if-nature-like-corporations-had-the-rights-

and-protections-of-a-person-64947. Accessed 22 July 2021. 

Culler, Jonathan. “Semiotics as a Theory of Reading.” The Pursuit of Signs: Semiotics, Literature, Deconstruction, 

Cornell University Press, 1981. 47-51. Print. 

“Dada.” Poetry Foundation, Poetry Foundation, https://www.poetryfoundation.org/learn/glossary-

terms/dada. Accessed 22 July 2021. 

Delvin, Es. Poem Portraits, Google Arts & Culture, https://artsexperiments.withgoogle.com/poemportraits. 

Accessed 22 July 2021. 

http://www.acrosticpoem.org/
http://www.ubu.com/aspen/aspen5and6/threeEssays.html
mailto:medium.com/@yuribarzov/humanness-is-underrated-d2b745124dd4
https://towardsdatascience.com/manifeto-time-for-humanness-learning-51051befa5f4
http://www.digitalhumanities.org/dhq/vol/11/3/000325/000325.html
http://www.nickbostrom.com/papers/history.pdf
https://www.writing.upenn.edu/~afilreis/88v/burroughs-cutup.html
https://www.writing.upenn.edu/~afilreis/88v/burroughs-cutup.html
https://carletonu.pressbooks.pub/digh5000/
https://www.computationalcreativity.net/proceedings/ICCC-2012-Proceedings.pdf
https://theconversation.com/what-if-nature-like-corporations-had-the-rights-and-protections-of-a-person-64947
https://theconversation.com/what-if-nature-like-corporations-had-the-rights-and-protections-of-a-person-64947


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  23 

Dormehl, Luke. “Truly creative A.I. is just around the corner. Here’s why that’s a big deal.” Digital Trends 

Media Group, January 5, 2018, www.digitaltrends.com/cool-tech/artificial-intelligence-creativity-future/. 

Accessed 22 July 2021. 

Eliot, T. S. (Thomas Stearns). The Sacred Wood: Essays on Poetry and Criticism, London: Methuen, 1928. Print. 

European Parliament Research Service. The ethics of artificial intelligence: Issues and initiatives. European Parliament, 

2020, 

https://www.europarl.europa.eu/RegData/etudes/STUD/2020/634452/EPRS_STU(2020)634452_EN.

pdf. Accessed 22 July 2021. 

“Flarf.” Poetry Foundation, Poetry Foundation, https://www.poetryfoundation.org/learn/glossary-

terms/flarf. Accessed 22 July 2021. 

Garcia, Robert. “Artificial Intelligence and Personhood.” Cutting Edge Bioethics: A Christian Exploration of 

Technology and Trends. Eds. John Kilner et al. Grand Rapids, MI: W.B. Eerdmans, 2002. Print.  

Gervás, Pablo. “Exploring Quantitative Evaluation of the Creativity of Automatic Poets.” Computational 

Creativity: The Philosophy and Engineering of Autonomously Creative Systems. Eds. Tony Veale et al. Cham:  

Springer, 2019. 275-304. Internet resource. 

Gervás, Pablo. “Generating poetry from a prose text. Creativity versus faithfulness.” 2001a, 

www.researchgate.net/publication/2925641_Generating_Poetry_from_a_Prose_Text_Creativity_versus_

Faithfulness. Accessed 22 July 2021. 

Gervás, Pablo. “An expert system for the composition of formal Spanish poetry.” Journal of Knowledge-Based 

Systems, vol. 14, no. 3-4, 2001. 181-188. 

https://www.sciencedirect.com/science/article/abs/pii/S0950705101000958. Accessed 22 July 2021. 

Glatch, Sean, et al. “What Is Blackout Poetry? Examples and Inspiration.” Writers.com, 11 June 2021, 

https://writers.com/what-is-blackout-poetry-examples-and-inspiration. Accessed 22 July 2021. 

Gnoetry Daily, https://gnoetrydaily.wordpress.com/. Accessed 22 July 2021. 

Gonçalo Oliveira, Hugo. “A Survey on Intelligent Poetry Generation: Languages, Features, Techniques, 

Reutilisation and Evaluation.” Proceedings of the 10th International Natural Language Generation conference, INLG 

201 (Santiago de Compostela, Spain), Association for Computational Linguistics, 2017. 11-20. 

https://aclanthology.org/W17-3502/ Accessed 22 July 2021. 

Gonçalo Oliveira, Hugo, et al. “Multilanguage extension and evaluation of a poetry generator.” Natural 

Language Engineering, 2017, nil.fdi.ucm.es/sites/default/files/multilingual_poetry.pdf. Accessed 22 July 

2021. 

Gonçalo Oliveira, Hugo. “Tra-la-Lyrics 2.0: Automatic Generation of Song Lyrics on a Semantic Domain.” 

Journal of Artificial General Intelligence, vol. 6, no. 1, 2015. 

https://www.researchgate.net/publication/291386243_Tra-la-

Lyrics_20_Automatic_Generation_of_Song_Lyrics_on_a_Semantic_Domain. Accessed 22 July 2021. 

Gonçalo Oliveira, Hugo. “Poetryme.” PoeTryMe, Center of Informatics and Systems of the University of 

Coimbra, 2015, https://poetryme.dei.uc.pt/ . Accessed 22 July 2021. 

http://www.digitaltrends.com/cool-tech/artificial-intelligence-creativity-future/
https://www.europarl.europa.eu/RegData/etudes/STUD/2020/634452/EPRS_STU(2020)634452_EN.pdf
https://www.europarl.europa.eu/RegData/etudes/STUD/2020/634452/EPRS_STU(2020)634452_EN.pdf
http://www.researchgate.net/publication/2925641_Generating_Poetry_from_a_Prose_Text_Creativity_versus_Faithfulness
http://www.researchgate.net/publication/2925641_Generating_Poetry_from_a_Prose_Text_Creativity_versus_Faithfulness
https://www.sciencedirect.com/science/article/abs/pii/S0950705101000958
https://writers.com/what-is-blackout-poetry-examples-and-inspiration
https://gnoetrydaily.wordpress.com/
https://aclanthology.org/W17-3502/
http://nil.fdi.ucm.es/sites/default/files/multilingual_poetry.pdf
https://www.researchgate.net/publication/291386243_Tra-la-Lyrics_20_Automatic_Generation_of_Song_Lyrics_on_a_Semantic_Domain
https://www.researchgate.net/publication/291386243_Tra-la-Lyrics_20_Automatic_Generation_of_Song_Lyrics_on_a_Semantic_Domain
https://poetryme.dei.uc.pt/


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  24 

Gonçalo Oliveira, Hugo. “PoeTryMe: a versatile platform for poetry generation.” Proceedings of ECAI 2012 

Workshop on Computational Creativity, Concept Invention, and General Intelligence, Montpellier, France, C3GI 

2012, 2012. https://eden.dei.uc.pt/~hroliv/pubs/GoncaloOliveira2012_c3gi_CRC.pdf. Accessed 22 July 

2021. 

Gonçalo Oliveira, Hugo, et al. “An Approach to Generate Text based on Rhythm.” New Trends in Artificial 

Intelligence. Eds. José Neves et al. Guimares, Portugal: Associação Portuguesa para a Inteligência Artificia, 

2007. 57-68. 

“Google’s AI Writes Poetry after Reading Thousands of Romance Novels.” Writers Write, Writers Write, Inc., 

19 May 2016, https://www.writerswrite.com/google-ai-writes-poetry-51920161. Accessed 22 July 2021. 

Gosper, R. W. “Item 176.” HAKMEM, Memo 239. Eds. Michael Beeler et al. Cambridge: Artificial 

Intelligence Laboratory, Massachusetts Institute of Technology, 1972. Internet resource. 

Gunkel, David and Wales, Jordan Joseph. “Debate: What is personhood in the age of AI?” AI & Society, vol. 

36, 2021: 473-486. Print. 

“Haiku Generator.” FantasyNameGenerators.com, https://www.fantasynamegenerators.com/haiku-

generator.php. Accessed 22 July 2021. 

Harel, David  and Rumpe, Bernhard. “Modelling Languages: Syntax, Semantics and All That Stuff – Part I: 

The Basic Stuff.” 2003, 

https://www.researchgate.net/publication/2877092_Modeling_Languages_Syntax_Semantics_and_All_

That_Stuff_Part_I_The_Basic_Stuff. Accessed 22 July 2021. 

Holland, Norman N. “Reading and Identity: A Psychoanalytic Revolution.” Academy Forum, American 

Academy of Psychoanalysis, vol. 23, 1979: 7-9. Print. 

Hrešková, Miroslava and Machová, Kristína. “Haiku poetry generation using interactive evolution vs. poem 

models.”Acta Electrotechnica et Inforamatica, vol. 17, no. 1, 2017: 10-16. DOI:10.15546/aeei-2017-0002 

Henrickson, Leah. humanities computing, digital humanities, and computational humanities: what’s in a name. 3:AM 

Magazine, October 24, 2019. https://www.3ammagazine.com/3am/humanities-computing-digital-

humanities-and-computational-humanities-whats-in-a-name/. Accessed 22 July 2021. 

Iser, Wolfgang. “Indeterminacy and the Reader’s Response in Prose Fiction.” Aspects of Narrative (English 

Institute Essays). Ed. Joseph Hillis Miller.  New York : Columbia University Press, 1971. 1-45. Print. 

James, William. The Principles of Psychology. New York: Henry Holt and Company, 1890. Print. 

Janes, Tyler L. “Legal personhood for Artificial Intelligence: Citizenship as the exception to the rule.” AI and 

Society, vol. 35, no. 2, 2020: 343-354. Print. 

Jauss, Hans R. “Literary History as a Challenge to Literary Theory.” Toward an Aesthetic of Reception. Transl. 

Timoty Bahti. Minneapolis: University of Minessota Press, 1967/1982. 3-45. Print. 

Kant, I. The Moral Law, translation of groundwork for the metaphysics of morals. London : Hutchinson University 

Library, 1972. Print. 

Kapell, Dave. Magnetic Poetry: Magnets that Reveal the Poet Within. https://magneticpoetry.com/. 

Accessed 22 July 2021. 

https://eden.dei.uc.pt/~hroliv/pubs/GoncaloOliveira2012_c3gi_CRC.pdf
https://www.writerswrite.com/google-ai-writes-poetry-51920161
https://www.fantasynamegenerators.com/haiku-generator.php
https://www.fantasynamegenerators.com/haiku-generator.php
https://www.researchgate.net/publication/2877092_Modeling_Languages_Syntax_Semantics_and_All_That_Stuff_Part_I_The_Basic_Stuff
https://www.researchgate.net/publication/2877092_Modeling_Languages_Syntax_Semantics_and_All_That_Stuff_Part_I_The_Basic_Stuff
https://www.3ammagazine.com/3am/humanities-computing-digital-humanities-and-computational-humanities-whats-in-a-name/
https://www.3ammagazine.com/3am/humanities-computing-digital-humanities-and-computational-humanities-whats-in-a-name/
https://magneticpoetry.com/


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  25 

Kapell, Dave, and Sally Steenland. The Magnetic Poetry Book of Poetry. New York: Workman Pub, 1997. Print.  

Koeneke, Rodney. “bug’s psalm.” https://www.poetryoutloud.org/poem/bugs-psalm/. Accessed 22 July 

2021. 

Lamb, Carolyn, et al. “A Taxonomy of Generative Poetry Techniques.” In BRIDGES Finland Conference 

Proceedings, 2017. 195-202. https://archive.bridgesmathart.org/2016/bridges2016-195.pdf. Accessed 22 

July 2021. 

Lamb, Carolyn, et al. “Can human assistance improve a computational poet?” In BRIDGES Finland Conference 

Proceedings, 2015. 37-44. http://t.archive.bridgesmathart.org/2015/bridges2015-37.pdf. Accessed 22 July 

2021. 

Lau, Jey Han et al. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: 

Long Papers). Melbourne: Association for Computational Linguistics, 2018. 

https://aclanthology.org/P18-1181and https://aclanthology.org/P18-1181.pdf. Accessed 22 July 2021. 

Laws, Amy E. Rimbaudelaires. ELMCIP, https://elmcip.net/creative-work/rimbaudelaires. Accessed 22 July 

2021. 

Levy, Robert P. “Rplevy/Poevolve: A Naive Model of Human Creativity / Dadaistic Song-Writing Code.” 

POEVOLVE, GitHub, https://github.com/rplevy/poevolve. Accessed 22 July 2021. 

Lutz, Theo. “Stochastische Texte.”  augenblick 4 (1959), H. 1, S. 3-9. https://www.stuttgarter-

schule.de/lutz_schule_en.htm . Accessed 22 July 2021. 

Mathews, Harry. 15-line snowball poem. https://poetrywithmathematics.blogspot.com/2010/08/snowballs-

growingshrinking-lines.html. Accessed 22 July 2021. 

Manurung, Hisar. An evolutionary algorithm approach to poetry generation. Ph.D. Dissertation. University of 

Edinburgh, 2004. Computer file. 

Manurung, Hisar. “A chart generator for rhythm patterned text.” Proceedings of the First International Workshop on 

Literature in Cognition and Computer (Tokyo), 1999. Internet resource. 

Matthias, Andreas. “The responsibility gap: Ascribing responsibility for the actions of learning automata.” 

Ethics and Information Technology, vol. 6, no. 3, 2004: 175-183. https://doi-

org.libproxy.uoregon.edu/10.1007/s10676-004-3422-1. Accessed 22 July 2021. 

Medvedev, Pavel Nikolaevich and Bakhtin, Mikhail Mikhaĭlovich. “The Object, Tasks and Methods of 

Literary History.” The Formal Method in Literary Scholarship: A Critical Introduction to Sociological Poetics. Transl. 

A.J. Wehrle. Baltimore: Johns Hopkins University Press, 1978. 26-37. Print. 

Miller, E. Ce. “Blackout Poetry Is A Fascinating Art Form You Can Try At Home Right Now.” August 28, 

2017, https://www.bustle.com/p/what-is-blackout-poetry-these-fascinating-poems-are-created-from-

existing-art-78781. Accessed 22 July 2021. 

Netzer, Yael, et al. “Gaiku: Generating haiku with word associations norms.” Proceedings of the Workshop on 

Computational Approaches to Linguistic Creativity, Association of Computational Linguistics, 2009. 32-39. 

Internet resource. 

Open Letter to the European Commission Artificial Intelligence and Robotics, http://www.robotics-

openletter.eu/. Accessed 22 July 2021. 

https://www.poetryoutloud.org/poem/bugs-psalm/
https://archive.bridgesmathart.org/2016/bridges2016-195.pdf
http://t.archive.bridgesmathart.org/2015/bridges2015-37.pdf
https://aclanthology.org/P18-1181
https://aclanthology.org/P18-1181.pdf
https://elmcip.net/creative-work/rimbaudelaires
https://github.com/rplevy/poevolve
https://www.stuttgarter-schule.de/lutz_schule_en.htm
https://www.stuttgarter-schule.de/lutz_schule_en.htm
https://poetrywithmathematics.blogspot.com/2010/08/snowballs-growingshrinking-lines.html
https://poetrywithmathematics.blogspot.com/2010/08/snowballs-growingshrinking-lines.html
https://doi-org.libproxy.uoregon.edu/10.1007/s10676-004-3422-1
https://doi-org.libproxy.uoregon.edu/10.1007/s10676-004-3422-1
https://www.bustle.com/p/what-is-blackout-poetry-these-fascinating-poems-are-created-from-existing-art-78781
https://www.bustle.com/p/what-is-blackout-poetry-these-fascinating-poems-are-created-from-existing-art-78781
http://www.robotics-openletter.eu/
http://www.robotics-openletter.eu/


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  26 

Oulipo (Ouvroir de Littérature Potentielle). Atlas de littérature potentielle, vol. 1 in Collection Idées. Paris: 

Gallimard, 1981. Print. 

Palmer, Michael. “Letter 7.” Poetry Foundation, Poetry Foundation, 

https://www.poetryfoundation.org/poems/51141/letter-7. Accessed 22 July 2021. 

Poem Generator, Powered by Aardgo, https://www.poem-generator.org.uk/. Accessed 22 July 2021. 

“Poetry.” Language Is a Virus, http://www.languageisavirus.com/poetry/index.php. Accessed 22 July 2021. 

Richards, Ivor Amstrong. Poetics and Sciences: A Reissue of “Science and Poetry” (1926, 1935) with Commentary. 

London: Routledge and Kegan Paul, 1970. Print. 

Roque, Antonio. “Language Technology Enables a Poetics of Interactive Generation.” Journal of Electronic 

Publishing, vol. 14, no. 2, 2011. https://doi.org/10.3998/3336451.0014.209. Accessed 22 July 2021. 

Schwatz, Oscar. “Can a Computer Write Poetry?” TEDx. TEDxYouth@Sydney, Sydney, Australia. 

https://www.ted.com/talks/oscar_schwartz_can_a_computer_write_poetry. Accessed 22 July 2021. 

Shihadeh, Juliana and Ackerman, Margaret. “EMILY: An Emily Dickinson machine.” Proceedings of the 11th 

International Conference on Computational Creativity (ICCC 2020), 2020. Internet resource. 

Stagnaro, Angelo. The Italian Jesuit Who Taught Computers to Talk to Us. January 28, 2017. 

https://www.ncregister.com/blog/the-italian-jesuit-who-taught-computers-to-talk-to-us. Accessed 22 

July 2021. 

Stein, Gertrude. “Red Faces”; “America.” https://allpoetry.com/Gertrude-Stein. Accessed 22 July 2021. 

Stephens, Wallace. “The Snow Man by Wallace Stevens.” Poets.org, Academy of American Poets, 

https://poets.org/poem/snow-man. Accessed 22 July 2021. 

“The Genuine Haiku Generator.” Everypoet.com, Selendy Communications, 

http://www.everypoet.com/haiku/. Accessed 22 July 2021. 

Tobing, Berty Chrismartin Lumban and Manurung, Ruli. “A chart generation system for topical metrical 

poetry.” Proceedings of the 6th International Conference on Computational Creativity (ICCC 2015), Park City, Utah, 

2015. Internet resource. 

Todorov, Tzvetan. Introduction to Poetics. Transl. Richard Howard. Brighton: Harvester Press, 1968/1981. Print. 

Tzara, Tristan, Dada Manifesto, 1918.  https://writing.upenn.edu/library/Tzara_Dada-Manifesto_1918.pdf. 

Accessed 22 July 2021. 

Tzara, Tristan. Three Poems by Tristan Tzara. Translate by Forrest Pelsue. 

https://exchanges.uiowa.edu/issues/cae-sura/three-dada-poems/. Accessed 22 July 2021. 

Veale, Tony. “Less rhyme, more reason: Knowledge-based poetry generation with feeling, insight and wit.” 

Proceedings of the 4th International Conference on Computational Creativity (ICC 2013), Sydney, Australia, 2013. 

152-159. Internet resource. 

Verse by Verse, Google AI Semantic Experiences, https://sites.research.google/versebyverse/. 

Accessed 22 July 2021.  

https://www.poetryfoundation.org/poems/51141/letter-7
https://www.poem-generator.org.uk/
http://www.languageisavirus.com/poetry/index.php
https://doi.org/10.3998/3336451.0014.209
https://www.ted.com/talks/oscar_schwartz_can_a_computer_write_poetry
https://www.ncregister.com/blog/the-italian-jesuit-who-taught-computers-to-talk-to-us
https://allpoetry.com/Gertrude-Stein
https://poets.org/poem/snow-man
http://www.everypoet.com/haiku/
https://writing.upenn.edu/library/Tzara_Dada-Manifesto_1918.pdf
https://exchanges.uiowa.edu/issues/cae-sura/three-dada-poems/
https://sites.research.google/versebyverse/


Humanist Studies & the Digital Age  Linardaki 

7.1 2022  27 

Winkler, Adam. “'Corporations Are People' Is Built on an Incredible 19th-Century Lie.” The Atlantic, Atlantic 

Media Company, 21 Apr. 2020, https://www.theatlantic.com/business/archive/2018/03/corporations-

people-adam-winkler/554852/. Accessed 22 July 2021. 

Wimmer, Mario. “Josephine Miles (1911–1985): Doing Digital Humanism with and without Machines.” 

History of Humanities, vol. 4, no. 2, 2019,  

www.journals.uchicago.edu/doi/abs/10.1086/704850?journalCode=hoh. Accessed 22 July 2021. 

Wong, Martin Tsan and Chun, Andy Hon Wai. “Automatic haiku generation using VSM.”  Proceedings of 7th 

WSEAS Int. Conf. on Applied Computer & Applied Computational Science (ACACOS ’08), Hangzhou, China, 

2008. Internet resource. 

Wright, George A. and Purver, Matthew. “Creative language generation in a society of engagement and 

reflection.” Proceedings of the 11th International Conference on Computational Creativity (ICC 2020), Coimbra, 

Portugal, 2020. 169-172. Internet resource. 

Yampolskiy, Roman V. “Could an Artificial Intelligence Be Considered a Person under the Law?” PBS, Public 

Broadcasting Service, 7 Oct. 2018, https://www.pbs.org/newshour/science/could-an-artificial-

intelligence-be-considered-a-person-under-the-law. Accessed 22 July 2021. 

https://www.theatlantic.com/business/archive/2018/03/corporations-people-adam-winkler/554852/
https://www.theatlantic.com/business/archive/2018/03/corporations-people-adam-winkler/554852/
http://www.journals.uchicago.edu/doi/abs/10.1086/704850?journalCode=hoh
https://www.pbs.org/newshour/science/could-an-artificial-intelligence-be-considered-a-person-under-the-law
https://www.pbs.org/newshour/science/could-an-artificial-intelligence-be-considered-a-person-under-the-law

	Poetry at the first steps of Artificial Intelligence
	Christina Linardaki
	Introduction
	Why poetry in the first place?
	The baby steps of poetry generation
	Digital Humanities vs Computational Creativity
	Online poetry generators for fun
	Poetry generators as academic projects
	Methods of mere poetry generation
	Enhancement of poetry generation results
	Human enhancement
	Computer enhancement
	Poetry generation in terms of target features
	Form features
	Content features
	Human or Not
	Dada
	Stream of consciousness
	OuLiPo
	Language Poetry
	Flarf
	Blackout or Erasure Poetry
	Human or Bot
	Works Cited


