Who's Talking: Dada, Machine Writing, and the Found

RiTa.js for Visual Artists and Writers

kathy wu
Brown Literary Arts

This assignment asks undergraduate students to generate text using both analog cut up techniques, as well as a simple Markov procedure, and discuss the power relations inherent in found writing processes. Through reading and making, students will encounter and critically develop their own articulations of found art—its questions of property and power—and how it relates to generative text and its corpuses.

Learning Goals: 

  • Gain a basic familiarity with generative text models, specifically Markov chains
  • Articulate points of view on the ethics of authorship within found text, specifically in the context of dadaism and machine writing
  • Produce a cross-disciplinary work with emphasis on writing, aesthetics, and computation.

Original Assignment Context: Middle of elective graphic design course

Materials Needed: Selected readings, RiTA.js markov library (web-accessible, see more details below), a free account on glitch.com

Time Frame: ~2-3 weeks


The following is an assignment for artists and writers to work with found materials, first via analog process, and then via the RiTa.js markov library, which uses a probabilistic model to generate text in the browser. The assignment relies on critical reading and making to ask, To what degree is all generative text engaging with found material, taken or given? 

This is intended to be completed over 2-3 course sessions. The instructor can also modify this assignment and use only RiTa.js (Part II) alongside the readings. 

I taught a modified version of this assignment for a course called Computational Poetics at RISD in 2021, both in the spring and fall. The course was listed under Graphic Design, and thus had a strong emphasis on typography and communication mediums. In both courses, the class comprised mostly undergraduates in Graphic Design, but also included many other majors across fine arts, as well as graduates across the school.

We began with cut-up poetry – Tristan Tzara’s “To Write a Dadaist Poem,” using text from our own surroundings. Students are invited to practice alongside Ron Padgett's Creative Reading, a text which offers playful, practical guidance for deriving new texts from existing ones. 

The class also discussed Kenneth Goldsmith’s notion of “uncreative writing,” which boldly encourages repurposing in the age of the internet, and—at first glance—seems to open up possibilities outside traditional definitions of plagiarism. Students, however, were quick to question Goldsmith: What does it mean for him, a white avant-garde artist, to recite whatever he wants as “found,” as “uncreative writing”? 

This question is well-framed by Eunsong Kim’s essay, “FOUND, FOUND, FOUND; LIVED, LIVED, LIVED,” which criticizes Goldsmith’s “found” performance of Michael Brown’s autopsy in 2014; which questions any institution’s claim over a “found” archival memory; which provides a lyrical and smart framework for discussing power, form, and content. I would encourage instructors to center this text in the workshop, and for students to read it thoughtfully before creating their own texts.

Instructors may want to assign Kim’s criticism alongside Robin Coste Lewis’s Voyage of the Sable Venus, which repurposes museum descriptions of the Black female figure to underscore institutional constructs of property, gender, and race. Lillian-Yvonne Bertram’s Travesty Generator is another useful reference, which iterates systemic racial violence via open source Python—like an ouroboros, at times the poems metabolize their own outputs towards deeply affecting ends. 

The goal, I emphasized, is not to write off “the found” entirely, but as Kim puts it, to reject this particular kind of found:

“The found that declares MINE when movements are in place tending to the damage. The found that declares MINE to be property, property without memory, property for sale. We care not one bit about: conceptualism, conceptualist strategies, the branding, the legacy, the tradition, the threat it supposedly “poses” against the equally omnipresent white lyric I (and what does it mean that advocates of the “I” and opposers of the “I” cannot and refuse to discuss the relationship between power and language, whiteness and language?). We find the language of both notions to be dull, rooted in the imagination of capital. We do not believe that form and content are ever separable.”

For the technical part of this unit, we covered basic knowledge of Markov chains and n-grams, which are a stochastic model for text generation that is “memoryless,” or in other words– it does not use past generation to infer future generation. We discussed the following questions: 

  • Who is Markov, and how might this technology be used? 
  • How might this feel aesthetically similar to something such as, say, your phone’s autocomplete?
  • What poetic possibility is left open in Markov chains’ “memorylessness”?
  • And finally, how might this relate to Kim’s “found devoid of memory”? This last question has yet to be probed, but I would be curious what other instructors find in this.

We used Glitch to make our own websites in combination with RiTa.js. RiTa.js is a library created by Daniel Howe for relative beginners to do quick and satisfying things using natural language processing in the browser. With just a few lines of code, we were able to generate 1000+ words using texts from Project Gutenberg. 

When playing with a demo of RiTa.markov() within the text generation, students were attentive to moments where other registers crept in: the unedited legal redistribution notice from Project Gutenberg, for instance. In a more explicit example on the RiTa.js website, two texts are combined: writings from both Kafka and Wittgenstein. The resulting output is a strange blend of both, where seams are not always obvious. 

In these moments, I encouraged students to think about how form talks to content, how it might distort or amplify it. What would have been helpful in these explorations is to share different visual treatments of text? One nice, simple example is Pamela Mishkin’s writing with GPT3, on Love and AI. But can one go further? How might students as artists and writers render the computer voice(s), the human voice(s), “found” textual objects to reveal their provenance, or not?

This assignment is designed for students with an existing art or writing practice, as well as students with beginning exposure to HTML, but who may or may not have worked with javascript or code, but can be adapted to students of a higher coding level as well.

The learning goals for the assignment span computational skills and humanistic questions. Through this assignment, students will gain a basic familiarity with generative text models, specifically Markov chains. They should be able to articulate their own points of view on the ethics of authorship within found text, specifically in the context of dadaism and machine writing. They will also have produced a cross-disciplinary work with emphasis on writing, aesthetics, and computation.

For this assignment, instructors and students will need a computer and internet access, as well as a beginner-friendly text editing software or an account on Glitch. Glitch was useful in my studio classroom because it allows beginners to quickly publish, share, copy, and collaborate on websites at no cost. Students can also upload text files and images. For source texts, I recommend Project Gutenberg as a starting point; students can also use plain text files which they have created themselves.

The Assignment

The assignment consists of three parts, outlined as followed:

We reject the notion of a scientific found. Of the removed found. Of the found that does not live. Of the found that institutions practice. Of found devoid of memory.
– Eunsong Kim

PART I: Found Poems

image src: https://www.brown.edu/Departments/Italian_Studies/n2k/multiplicity/JBarret/block10.html

In class together, we’ll look at:

  • Tristan Tzara’s To Write a Dadaist Poem
  • OuLiPo
  • Ron Padgett's Creative Reading

After class, find sources of text (receipts, pages of books, street signs, text on the internet), and scan or photograph it, then reprint it. Cut this up into discrete pieces. You can use work with images as well. Collage this and then rescan it. Produce 2-3 pages from this exercise.

Read and take notes on FOUND, FOUND, FOUND; LIVED, LIVED, LIVED by Eunsong Kim. Optionally, also read excerpts from The Sable Venus by Robin Coste Lewis and Travesty Generator by Lillian-Yvonne Bertram.

PART II: Markov Memory

In class, we will be using RiTa.markov() with RiTa.js. You will define the “text” variable as whatever you’d like; “text” should be data that is in a .txt file, for example. 

Try replacing the templated text with your own text.

function generate(){
        // create a markov model w' n=4
        markov = RiTa.markov(4);
        // load text into the model
        //generate 10 lines
        genText = markov.generate(10);
        //adds a space between all the random stuff
        //add it to the HTML
        document.getElementById("container").innerHTML = genText;

After class, “write” 3-5 pages of poetry or prose or language art using RiTa markov and at least two sources of text. In the first piece, generate a model with one existing text. This can be a book from Project Gutenberg, or an article, anything which can be copy and pasted as plaintext. In the second text, include the previous pieces of text but include your own writing within the model. 

Consider how memory functions in your work. You might consider beginning your writing with “Remembering is…” or “Forgetting is…” What does the computer “know,” and “remember”? What does it forget? 

Print this work and bring it to class. Include annotations and observations you have in the margins of your submission.

PART III: Form & Content

For the final project, present your combined new texts into a final bound book or website, or another medium or your choice. Write a 200 word reflection, to turn this in alongside your work.

  • What formal choices are you making to clarify the content? Color, typography?
  • Where are areas of slippage or strangeness in the generation? What are its limits, and where does it create the unexpected?
  • Who is the author of this work? Who takes credit? Who is present in the piece? What other voices might be erased or extraneously present? Use a quote from Kim to frame your thinking.


Kim, Eunsong. “FOUND, FOUND, FOUND; LIVED, LIVED, LIVED.” http://www.scapegoatjournal.org/docs/09/Eunsong%20KIM,%20Found,%20Found,%20Found_%20Lived,%20Lived,%20Lived.pdf. Scapegoat Journal 09, EROS. Accessed February 11, 2023. 

Parrish, Allison. “N-Grams and Markov Chains.” decontextualize, March 11, 2014. https://www.decontextualize.com/teaching/rwet/n-grams-and-markov-chains/.

Staff, Harriet. “Robin Coste Lewis's Voyage of the Sable Venus Reviewed at The Rumpus.” Poetry Foundation. Poetry Foundation, November 10, 2015. https://www.poetryfoundation.org/harriet-books/2015/11/robin-coste-lewiss-voyage-of-the-sable-venus-reviewed-at-the-rumpus.

Howe, Daniel. “Tutorial: Generating with N-Grams.” Observable, May 1, 2022. https://observablehq.com/@dhowe/tut-rita-ngrams.

Bertram, Lillian-Yvonne. Travesty Generator. Noemi Press, 2019. 

Padgett, Ron. Creative Reading: What It Is, How to Do It, and Why. National Council of Teachers of English, 1997.


Markov example: https://glitch.com/edit/#!/rita-markov

Original course syllabus for Computation Poetics: https://kaaathy.com/comppoetics/

Project Gutenberg: https://www.gutenberg.org/

“How to Make a Dadaist Poem” in text format: https://www.writing.upenn.edu/~afilreis/88v/tzara.html

Example of Pamela Mishkin’s writing with GPT3: https://pudding.cool/2021/03/love-and-ai/. Licensed under Creative Commons Attribution-NonCommercial.