Poem Machine: Krock/Crash, 2026

Algorithmic poetry / Typography / Processing / video, 6:58
Seven generated poems

Poem Machine: Krock/Crash is the latest iteration of Poem Machine, an ongoing algorithmic poetry project in which I use code simultaneously as a writing system, typographic apparatus, visual medium and performance instrument. The project is iteration-oriented rather than directed towards a definitive version. Its linguistic corpus, typography, interfaces and modes of presentation have changed according to place, collaborators and technical circumstances. One version may function primarily as projected generative poetry, another as a live audiovisual instrument, another as a multilingual printing system, and another as a collaborative structure through which several writers' vocabularies can occupy the same computational space.

Krock/Crash brings several of these trajectories together.
The title uses the Swedish krock, a collision or crash. The work consists of seven poems recorded from the running system. They are not seven texts written in advance and subsequently animated. They are seven temporary configurations produced by a live generative program whose vocabulary is finite but whose combinations are not predetermined.

From Design by Numbers to Processing

My relationship to creative coding predates Poem Machine. I have written generative code since the 1990s, but one important route into the visual programming environment I use today came through John Maeda's Design by Numbers.

I was trying to work with Design by Numbers when that path led me towards Processing. This connection is not incidental. Processing itself grew directly from Maeda's Design By Numbers project. Ben Fry and Casey Reas began Processing in 2001 while working within Maeda's Aesthetics and Computation group at the MIT Media Lab. They conceived it as a code-based sketchbook through which artists and designers could approach programming as an ordinary part of visual practice.

That lineage remains important to me because it shifts programming away from the idea that code belongs exclusively to software engineering. Maeda's Design by Numbers, and later Processing, suggested that a line of code could occupy a position analogous to a drawn line, a typographic decision or a compositional rule.

My own library contains Maeda's Design by Numbers, Creative Code and The Laws of Simplicity, alongside Reas and Fry's Processing: A Programming Handbook for Visual Designers and Artists. That genealogy matters because Poem Machine is not an application to which I later added an aesthetic layer. The code itself determines how language enters time, movement and form.

Processing remains particularly suited to this approach because it sits between programming and sketching. It gives me access to Java while retaining a relatively immediate visual environment. A fragment of code can be altered, run, observed and altered again. The result is a form of iterative composition in which programming decisions and visual decisions remain very close to one another. This position also belongs to a longer history of computational art. White Heat Cold Logic, edited by Paul Brown, Charlie Gere, Nicholas Lambert and Catherine Mason, documents British computer artists of the 1960s and 1970s who approached computation as an artistic medium, often learning programming, constructing systems and working alongside scientists instead of treating the computer as a neutral production tool. My interest in that history is less nostalgic than methodological. The important proposition is that the artist must engage with the logic of the machine sufficiently deeply for that logic to become part of the work.

Paul Fishwick's Aesthetic Computing similarly proposes that aesthetic thinking need not simply decorate computation after the fact. Art and aesthetics can alter how computation itself is conceptualised and represented. Donald Knuth's earlier formulation of “computer programming as an art”, and Pierre Lévy's De la programmation considérée comme un des beaux-arts, belong to the same wider argument in my reading: programming can be regarded not only as an instrumental procedure but as a cultural and compositional activity.

Geomerative: when a letter stops being only a letter

Since 2011, a particularly important tool in my Processing work has been Geomerative, created by Ricard Marxer, with Mark Luffel credited as a collaborator.

Geomerative is a Processing library for generative two-dimensional geometry. Among its most useful capabilities is its ability to interpret TrueType fonts and expose typography as vector geometry. Instead of receiving only a finished glyph from a text-rendering function, the programmer can access the paths, points and control handles from which that glyph is formed. This fundamentally changes my relationship to typography.

If I ask an ordinary text renderer to display the word language, it returns a recognisable typographic object. The internal construction of those letters remains largely hidden. With Geomerative, I can work before that finished condition. A glyph becomes a group of contours. A contour contains paths. Paths can be sampled as points. Those points can be moved, scattered, attracted towards target positions, disturbed by procedural noise or allowed to drift away again. A word therefore becomes simultaneously linguistic material and geometry. This is central to Poem Machine. I am not primarily applying animation to typography. I am manipulating the conditions under which typography becomes recognisable.

The viewer initially encounters marks, points and fragments whose status as language may be uncertain. Gradually, those marks approach their target coordinates. Letterforms become perceptible. Letters form words, and words eventually stabilise into a poem. That stable moment lasts only temporarily before the geometry begins to lose its organisation again. Readability has a duration. The poem does not simply appear and disappear. It passes through states of becoming-readable and becoming-unreadable.

Akzidenz-Grotesk, accident and the geometry of neutrality

For the Latin-script component of Poem Machine, I use Akzidenz-Grotesk Pro. My attraction to Akzidenz-Grotesk is partly formal. Its relatively direct grotesque construction provides a strong structural basis when transformed into paths and points. It does not need decorative characteristics to assert itself. When its contours are dispersed by the program, the underlying architecture of the letter remains sufficiently distinct for the transition between abstraction and legibility to become visible.

Its history adds another layer. The name Akzidenz-Grotesk can be translated approximately as “jobbing sans”. Akzidenzen referred to everyday commercial printing such as business cards, invoices and letterheads. More unexpectedly, Akzidenz is rooted in the Latin accidentia, referring to chance, casual events or things that happen. I do not claim that this etymology explains the historical design of the typeface, nor that chance was an intention of its designers. For Poem Machine, however, the coincidence is difficult to ignore.

A typeface associated with routine commercial printing becomes the carrier of algorithmic chance. Later, the same poems leave the screen through a thermal printer whose paper evokes receipts, tickets and transactional records. The everyday typography of Akzidenzen returns through a computational system governed by controlled accident. The connection is therefore both historical and material: jobbing print, accident, algorithmic recombination and receipt-like thermal paper.

The sans serif, often associated with typographic clarity and functional communication, is also made dysfunctional. Its points disperse. Its contours refuse to remain fixed. Legibility becomes contingent. This connects Poem Machine with a longer interest in dysfunction and the creative consequences of software error that I examined in my doctoral research, Aesthetics of Dysfunction: On Virtual Agglomerations and the Creative Errancies of Code. Dysfunction, in this sense, is not simply technological failure. It can be a method for exposing assumptions embedded in apparently stable systems.

Movement, easing and controlled noise

The movement of Poem Machine is deliberately different from conventional kinetic typography. Much kinetic typography begins with an already completed typographic object. A word or sentence exists in a readable state and is subsequently moved, enlarged, rotated, faded or otherwise animated. Poem Machine acts earlier. The letterform itself begins in a displaced condition. Points that eventually constitute the glyph are elsewhere. Their movement towards the correct typographic configuration produces the transition into readability. Easing controls the rate of this movement. A computer can move an object from coordinate A to coordinate B at a perfectly constant speed, but that produces an obviously mechanical transition. Easing changes the acceleration and deceleration over time. A form can move quickly and then settle, approach gradually, hesitate or appear to gather itself into position.

Controlled noise adds another form of instability. The system uses coherent procedural variation, including Processing's noise functions, to disturb position and motion without producing completely unrelated random values. Instead of static or arbitrary chaos, the movement can resemble vibration, drift or turbulence. The word controlled matters. If the disturbance is too small, the system becomes conventional typesetting with motion added afterwards. If it becomes too large, the viewer loses the possibility of reading altogether. The aesthetic condition I am interested in lies between those states. The viewer watches language struggle towards legibility. That process also exposes something that printed typography generally conceals. A printed word presents itself as stable. Computational typography reveals that every letter is ultimately a set of instructions, coordinates and rendering decisions. Poem Machine makes those instructions perceptible as behaviour.

How Poem Machine actually writes

Poem Machine is often described as generative poetry, but I think it is important to specify what “generative” means in this case. It is not generative AI. The work uses no Large Language Model, or LLM, no predictive text model and no external artificial intelligence service. The program cannot independently invent vocabulary. Everything it can say is constrained by material placed within its corpus and by rules written into the code.

The linguistic material is stored in UTF-8 text files. UTF-8 is a Unicode text encoding capable of representing characters from a very large number of writing systems. This becomes important once a supposedly single wordbank begins to contain Swedish, Bosnian, Icelandic, English and other material. The principal wordbanks are organised into categories such as subjects, verbs, adjectives, questions and endings.

The program then uses a programmed grammar to construct a seven-line poem. Some parts of a line come directly from those text files. Other parts are connective sentence structures written into the code. A subject might be inserted into one grammatical frame, an adjective into another, while a question or ending may function as a larger authored fragment. This is an important distinction. The software is not just shuffling complete poems, but neither is every word generated independently. Authorship is distributed across several layers: the source vocabularies, longer submitted fragments, connective phrases, sentence templates, selection rules, timing and live operation of the system. Processing performs pseudo-random selections within those constraints. The possible combinations can become extensive, but they remain bounded by what has been authored and programmed.

This places Poem Machine within a lineage of combinatorial literature. The obvious historical reference is Raymond Queneau's Cent mille milliards de poèmes of 1961. Queneau’s work consists of ten sonnets, each containing fourteen lines. The pages are cut horizontally so that every individual line can be turned independently of the others. This means that the first line of a poem can be chosen from any of the ten first lines, the second from any of the ten second lines, and so on through all fourteen positions. The number of possible combinations is therefore:

10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 × 10 = 10¹⁴

That is 100,000,000,000,000 possible sonnets, or, in French numbering, cent mille milliards de poèmes, literally “one hundred thousand billion poems”.

Crucially, Queneau did not simply cut unrelated lines into strips. He wrote the ten sonnets so that corresponding lines could be exchanged while preserving the sonnet’s metre, rhyme structure and grammatical coherence. The book is both a collection of poems and a combinatorial machine: a finite set of authored elements capable of producing an enormous number of possible texts.

Queneau's work remains important to me because the poem is not identified with one fixed sequence of words. The artwork includes the space of possible poems and the mechanism through which one of those possibilities becomes actual. There is nevertheless an important difference. Queneau's combinatorial system remains extraordinarily controlled. Every interchangeable line was written so that metre, rhyme and grammar would continue to operate. Poem Machine permits considerably more friction. A line may suddenly become strange. Syntax may momentarily fail. One image may contradict another. The result may sound deliberate despite having never previously existed. The machine does not know which of these outcomes is successful.It does not possess a semantic model capable of judging whether a line is poetic, absurd, politically charged, funny or unexpectedly moving. It generates conditions under which such judgements become possible for a human reader.

Human and machine

David Mindell's Digital Apollo: Human and Machine in Spaceflight has been useful to my thinking about this question. Mindell examines the relationship between astronauts, automation and the Apollo Guidance Computer, resisting the simple narrative in which increasing automation simply removes human agency. Apollo's systems instead required constantly negotiated relationships between automated computation and skilled human control.

Poem Machine operates in a very different context, but I find the distinction productive. The interesting question is not whether the machine or the human is “really” writing. The system consists of relationships between authored vocabularies, software constraints, pseudo-random selection, typographic behaviour, live performance and human interpretation. The computer does something that would be difficult for me to perform manually at the same speed and scale, but it never escapes the structures I have built for it. At the same time, those structures produce combinations I did not consciously anticipate. That is a more interesting condition than autonomy.

Hong Kong: Perpetual Prose and the move towards performance

The particular system that became Poem Machine: Perpetual Prose was developed in 2025 in preparation for my residency in Hong Kong through the Helsinki to Hong Kong Exchange Programme, part of Non-Local Non-Solo 2025. The programme was initiated, produced and organised by nProjekt, with Myymälä2 as its Helsinki partner and Pointsman Artspace and Mooroom as its Hong Kong venue collaborators, with support from the Hong Kong Arts Development Council. My residency culminated at Mooroom with Serendipity – A Non-Solo of Moe Louanjli in association with Heyse Ip, Yim Sui Fong and Gavin Yip. The Non-Solo model was conceived around questions of authorship, cultural exchange, translation and co-presence and not the isolation of a single visiting artist, placing the featured artist’s work in dialogue with locally based practices.

My Hong Kong exhibition, Serendipity – A Non-Solo of Moe Louanjli in association with Heyse Ip, Yim Sui Fong and Gavin Yip, was presented at Mooroom, 9/F, Cheong Tai Industrial Building, 16 Tai Yau Street, San Po Kong, Hong Kong, in November 2025. The exhibition ran from 16 to 29 November, with an artist talk and live performance on 15 November. Serendipity was curated by Jamsen Law, assisted by Lee Suet Ying and Kelvin Au.

The exhibition's Non-Solo structure became important for the subsequent development of Poem Machine. My work was placed in relation to Heyse Ip, Yim Sui Fong and Gavin Yip rather than presented as an isolated imported practice. Discussions around the exhibition, local language, sound, everyday material and the different artists' approaches acted as a catalyst for thinking about collaboration as something that might enter the internal structure of the software itself.

The question gradually shifted from “How can Poem Machine respond to a place?” to “What happens if other people's language enters the machine?”

Hong Kong and multiscript typography

Hong Kong also forced an important technical expansion. The original visual architecture of Poem Machine had been strongly oriented around Latin typography and Geomerative. For Hong Kong, I needed to work seriously with Traditional Chinese characters and Cantonese textual material.

The terminology matters. Cantonese is a language. Traditional Chinese refers to a form of written Chinese characters. CJK is a computing and typographic abbreviation for Chinese, Japanese and Korean. CJK is not a language in itself. It describes a family of script-related technical requirements arising partly from the extensive use of Han ideographs across these writing traditions.

For the Hong Kong version, I used Noto Sans HK, the Hong Kong variant within the Noto CJK family. Noto Sans HK is specifically designed for languages in Hong Kong that use Traditional Han forms. Another relevant acronym is HKSCS, the Hong Kong Supplementary Character Set, developed to encode characters required in the Hong Kong computing environment that were not adequately covered by earlier standard character sets. HKSCS is a character-encoding standard and not a typeface or a glyph design system.

Traditional Chinese should also not be described as an RTL, or right-to-left, script. Contemporary horizontal Chinese is generally set left-to-right, while traditional vertical composition runs top-to-bottom with successive columns progressing from right to left. RTL properly describes writing systems such as Arabic and Persian/Farsi, where the base direction of horizontal text is right-to-left. These differences eventually required Poem Machine to operate through more than one typographic rendering strategy.

Processing, Java and AWT

Processing's principal visual mode is built around Java, but Processing is not simply a synonym for Java. It provides a simplified environment designed around visual and interactive work while retaining access to the much larger Java ecosystem. Processing's own documentation describes it as originally being a domain-specific extension to Java for artists and designers. That access became crucial for multilingual typography.

For Latin text I could continue to use Geomerative, allowing letters to be treated as outlines and points. For scripts requiring more sophisticated shaping, glyph substitution or bidirectional behaviour, I turned to Java AWT. AWT means Abstract Window Toolkit. It is part of Java's standard desktop graphics architecture. Within it, the package java.awt.font provides classes concerned specifically with typography. One of the most important for Poem Machine is TextLayout.

TextLayout analyses styled character data and handles operations including font substitution, typographic metrics and bidirectional reordering. It can determine how characters should be positioned when a line contains left-to-right, right-to-left or mixed-direction material. This means the artwork effectively has two complementary typographic architectures. Geomerative allows me to violate the apparent stability of Latin typography by turning it into geometry. AWT allows me to preserve scripts whose internal typographic behaviour should not simply be reduced to the assumptions of Latin vector lettering. I do not consider this asymmetry a defect. On the contrary, forcing every writing system through one visual mechanism would create a false technical neutrality. Different scripts have different requirements. The software should acknowledge them.

The live instrument: MIDI, Organelle and synthesis
Hong Kong also transformed Poem Machine from a primarily generative display into something closer to a performable instrument.

For the live performance at Mooroom I used a Critter & Guitari Organelle M as a programmable sound-processing and MIDI-connected extension to the system, controlled in conjunction with a Korg monologue and an Arturia MiniLab MIDI controller. MIDI means Musical Instrument Digital Interface. MIDI does not normally transmit audio. It transmits control information such as note events, velocity values, controller changes, pitch bend and timing information.

The Organelle is particularly interesting because it is not simply a conventional synthesiser. It is an open programmable musical computer whose patches have traditionally been built in Pure Data, or Pd, a graphical programming environment for real-time sound. The Organelle can receive and transmit MIDI through both USB and TRS connections, allowing external controllers and synthesisers to become part of a larger computational system. The Korg monologue is a 25-key monophonic analogue synthesiser capable of transmitting MIDI information, while the Arturia MiniLab functions as a compact MIDI controller.

This configuration changed the status of Poem Machine.

The program no longer had to sit on a computer and autonomously generate visual poems. Parameters could be performed. MIDI Note On events could trigger actions. Continuous Controller values could affect behaviour. Pitch Bend and other signals could be mapped to aspects of the visual system. The system became an instrument. This is one reason I describe Poem Machine as iteration-oriented. Its identity does not depend upon one fixed hardware configuration. MIDI, microphones, synthesisers, printers and other inputs or outputs can be added according to the situation without the conceptual structure of the work disappearing.

From Hong Kong back to Helsinki
Shortly before the Hong Kong residency, in October 2025, I presented a Swedish iteration of Poem Machine: Perpetual Prose in Helsinki for Tidskriften Kontradiktion. It was around this presentation that Hilda Forss and Anders E Larsson proposed developing a more collaborative version of the project. The suggestion arrived at precisely the right point. The Non-Local Non-Solo experience in Hong Kong had already raised questions for me about relational structures and shared authorship. Instead of inserting local references into my own wordbanks, I became interested in what would happen if the linguistic source material itself belonged to other writers. That proposition became Poem Machine: Matchmaking.

Poem Machine: Matchmaking, Donner Residence, 25–29 May 2026
Poem Machine: Matchmaking was a working-week workshop that I facilitated at the Donner Residence in Kruununhaka, Helsinki, from Monday 25 May to Friday 29 May 2026, within the curatorial framework of Tidskriften Kontradiktion.

I worked with four poets: Madeleine Stijernberg, Monika Musovic Bosnic, Barış Yüksel and Gerd-Peter Löcke.

I worked with one poet during each working day. The purpose was not to ask each poet to write “computer poetry”. Instead, we considered how existing writing practices might enter a computational structure without being flattened into one generic voice. Each poet curated material that could be reorganised as subjects, adjectives, verbs, questions, endings and other fragments. This transformed the wordbank from something I had written for the machine into a site of collaboration. The working week concluded on Friday 29 May with a presentation-performance at the Donner Residence, simultaneously for an audience present in the room and through Zoom.

During that performance the poems were projected at architectural scale. Each poet's linguistic material passed through the visual system: geometry emerged, language became readable, remained briefly and dissolved again. A thermal printer produced poems on demand. Each poet could leave carrying a unique strip of generated text.

Thermal print: from event to trace
The thermal printer is important because it establishes another relationship to time. On screen the poem is luminous, animated and unstable. Its existence is measured in seconds. Printing makes one generated configuration physical, but thermal paper is not an archival solution in the traditional sense. It fades. It is vulnerable to heat, light and time. The material therefore preserves the logic of impermanence rather than cancelling it. The receipt-like form also returns unexpectedly to Akzidenz-Grotesk's historical association with jobbing print. The same system that converts Akzidenz letterforms into unstable geometry eventually produces a paper strip resembling a commercial machine record. I also use an optional UTC, or Coordinated Universal Time, signature inspired by ACARS, the Aircraft Communications Addressing and Reporting System. The timestamp records the moment at which a transient generated configuration becomes a physical object. The poem becomes an event record.

From Matchmaking to Krock/Crash
Krock/Crash takes the Matchmaking structure and removes the separation between the individual contributors. The four poets' corpora are combined. I enter the corpus as a fifth textual contributor, adding language drawn from the Donner week itself: workshop, collaboration, books, fika, the large table, wooden floor, projection light, thermal paper, radon meter, seafront, marina, sailing boats, Kruununhaka, Helsinki and Helsingfors. Some of my contribution consists of documentary nouns. Other elements become connective linguistic structures through which one contributor's material can encounter another's.

For the earlier multilingual versions of Poem Machine, the interface could select explicit language modes. Krock/Crash changes that architecture conceptually. All of the source material now enters through one active corpus trigger. The files may still carry the technical prefixes en_ but EN no longer means that the poem is English. It means: load this corpus. A single wordbank may contain several languages. Nothing needs to be translated because it has entered the same software structure. Swedish can remain Swedish. Bosnian can remain Bosnian. Icelandic can remain Icelandic. The machine treats linguistic difference as material more than a menu of national language categories. This is the point at which matchmaking becomes crash.

A marine image can meet a publishing meeting. Political language can encounter fika. Mythology can meet thermal paper. A radon meter can appear beside desire. One poet's sentence structure can suddenly carry another poet's image. A syntactic collision may fail. Another may become unexpectedly convincing. The computer cannot distinguish between them. The reader can.

Collision and constrained chance
The word crash risks suggesting only destruction. I am more interested in collision as a change of trajectory. Two systems encounter one another and neither necessarily leaves in the direction from which it arrived. That is why Mark Nunes's writing on error, glitch, noise and jam remains relevant to my thinking. An error does not have to be romanticised as automatically creative, but error can expose the invisible rules by which a system normally defines successful operation. The same concern runs through my earlier work on the aesthetics of dysfunction. Krock/Crash does not deliberately maximise nonsense. If every result were random, collision would lose significance. The work depends on a constrained grammar strong enough to produce recognisable language and loose enough to allow unexpected combinations. The crash occurs precisely because structures remain.

Software as cultural material
This is why I do not regard software as a neutral carrier for a pre-existing poem. Lev Manovich's Software Takes Command, Matthew Fuller's Software Studies, Rob Kitchin and Martin Dodge's Code/Space, Ed Finn's What Algorithms Want and Fishwick's Aesthetic Computing all approach computation, in different ways, as culturally consequential rather than merely technical. My bibliography contains these works alongside practical books on Processing, generative art and typography.
Finn's idea of the algorithm as something operating between idealised computational abstraction and messy cultural reality is especially pertinent. Algorithms do not remain in the purity of formal logic once they encounter actual language, actual bodies and actual institutions. Poem Machine makes that gap visible. Its grammar is formal. Its words are not. They carry memory, geography, politics, humour, biography and ambiguity into the system.

Krock/Crash, the video
The video Poem Machine: Krock/Crash lasts 6 minutes 58 seconds and contains seven generated poems. The video does not represent a definitive edition of those poems. It records seven states produced by the running system. Another execution can make another sequence. The animations have not been designed afterwards around predetermined sentences. The texts are assembled by the same generative process through which their letterforms become visible. This means that linguistic generation and visual generation belong to the same temporal event.

A word is selected. Its geometry is calculated. Its points move. The line acquires legibility. The poem holds. It dissolves. Another poem becomes possible. The code is simultaneously score, instrument, typography and performer. The human authors provide language. I construct the generative conditions and perform the apparatus. The machine executes the rules. The reader encounters the result. No one of these positions alone constitutes the poem.

Krock/Crash is curated by Tidskriften Kontradiktion, Hilda Forss and Anders E Larsson, for Tidskriften Kontradiktion #8. The vocabulary is finite. The encounters are not.

Selected bibliography
The following references form part of the wider artistic, technical and theoretical context of the project and are selected in substantial part from my working bibliography.

1. Bohnacker, Hartmut, Benedikt Gross, Julia Laub and Claudius Lazzeroni. Generative Design: Visualize, Program, and Create with Processing. New York: Princeton Architectural Press, 2012.

2. Brown, Paul, Charlie Gere, Nicholas Lambert and Catherine Mason, eds. White Heat Cold Logic: British Computer Art 1960–1980. Cambridge, MA: MIT Press, 2009.

3. Dyson, George. Turing’s Cathedral: The Origins of the Digital Universe. London: Allen Lane, 2012.

4. Finn, Ed. What Algorithms Want: Imagination in the Age of Computing. Cambridge, MA: MIT Press, 2017.

5. Fishwick, Paul A., ed. Aesthetic Computing. Cambridge, MA: MIT Press, 2006.

6. Fuller, Matthew, ed. Software Studies: A Lexicon. Cambridge, MA: MIT Press, 2008.

7. Knuth, Donald E. “Computer Programming as an Art.” Communications of the ACM 17, no. 12 (1974): 667–673.

8. Lévy, Pierre. De la programmation considérée comme un des beaux-arts. Paris: La Découverte, 1992.

9. Louanjli, Moe Mohamed. Aesthetics of Dysfunction: On Virtual Agglomerations and the Creative Errancies of Code. PhD thesis, Griffith University, 2013.

10. Lupton, Ellen. Thinking with Type: A Critical Guide for Designers, Writers, Editors, & Students. 2nd revised and expanded edition. New York: Princeton Architectural Press, 2010.

11. Maeda, John. Design by Numbers. Cambridge, MA: MIT Press, 2001.

12. Maeda, John. Creative Code. London: Thames & Hudson, 2004.

13. Maeda, John. The Laws of Simplicity. Cambridge, MA: MIT Press, 2006.

14. Manovich, Lev. The Language of New Media. Cambridge, MA: MIT Press, 2002.

15. Manovich, Lev. Software Takes Command. New York and London: Bloomsbury, 2013.

16. Mindell, David A. Digital Apollo: Human and Machine in Spaceflight. Cambridge, MA: MIT Press, 2008.

17. Nunes, Mark. Error: Glitch, Noise, and Jam in New Media Cultures. New York: Continuum, 2011.

18. Pearson, Matt. Generative Art: A Practical Guide Using Processing. Shelter Island, NY: Manning, 2011.

19. Pohlen, Joep. Letter Fountain: On Printing Types. 4th ed. Köln: Taschen, 2011.

20. Queneau, Raymond. Cent mille milliards de poèmes. Paris: Gallimard, 1961.

21. Reas, Casey and Ben Fry. Processing: A Programming Handbook for Visual Designers and Artists. Cambridge, MA: MIT Press, 2007.

22. Reas, Casey, Chandler McWilliams and Jeroen Barendse. Form+Code in Design, Art, and Architecture. New York: Princeton Architectural Press, 2010.

Poem Machine: Krock/Crash är ett algoritmiskt poesiverk byggt i Processing utifrån en gemensam flerspråkig korpus av Madeleine Stijernberg, Monika Musovic Bosnic, Barış Yüksel, Gerd-Peter Löcke och Moe Louanjli.

Verket utvecklades ur Poem Machine: Matchmaking, en veckolång workshop som jag ledde på Donner Residence i Kronohagen, Helsingfors, i maj 2026. Varje poet bidrog med ord, fraser, frågor och textfragment ur sitt eget skrivande. I Krock/Crash förs dessa separata bidrag samman i en gemensam generativ korpus. Jag går själv in i systemet som en femte bidragsgivare med material hämtat från workshopen: böcker, fika, det stora bordet, termopapper, radonmätaren, marinan, projektionsljuset, Kronohagen, Helsinki och Helsingfors.

Språken separeras eller översätts inte. Svenska, engelska, bosniska, isländska och andra fragment kan förekomma i samma dikt. Programmet väljer material ur avgränsade textfiler med subjekt, verb, adjektiv, frågor och avslutningar och kombinerar dem enligt uttryckliga regler i koden. Ingen generativ AI eller Large Language Model används.

Krocken sker mellan författare, språk, minnen, platser och semantiska register. En marin bild kan möta förlagsvärlden; politiskt språk kan kollidera med fika; mytologi kan möta termopapper. Maskinen försöker inte lösa dessa möten till en enda röst. Den fortsätter att kombinera dem på nytt.

Visuellt gör Geomerative det möjligt att behandla bokstavsformer som geometri. Konturerna sprids, driver och närmar sig gradvis läsbarhet innan de åter blir instabila. Akzidenz-Grotesk ger den latinska typografin dess struktur. Typsnittets historiska koppling till vardagligt tryck, tillsammans med den etymologiska resonansen mellan Akzidenz och accidentia, händelse eller slump, knyter an till den algoritmiska kombinationen och de kvittoliknande termoutskrifter som utvecklades under Matchmaking.

Videon dokumenterar sju tillfälliga tillstånd i det generativa systemet i realtid. Dikterna skrevs inte i förväg för att sedan animeras. En annan körning kan skapa en annan sekvens.

Krock/Crash behandlar kollisionen inte som katastrof utan som ett kompositionellt villkor: fem röster går in i ett system utan att behöva bli en.

Ordförrådet är ändligt. Mötena är det inte.

Curaterad av Tidskriften Kontradiktion
Hilda Forss och Anders E Larsson
För Tidskriften Kontradiktion #8
Helsingfors, 2026