Mark Dingemanse

@markdingemanse.net

3078 Followers 519 Following

Language, interaction, tech • papers https://markdingemanse.net • blog https://ideophone.org • more colourful on fedi https://scholar.social/@dingemansemark/ • POSSE: Publish on Own Site, Syndicate Everywhere

Highlight from conclusion of paper: There could hardly be a more effective way for a computational artifact to exploit human interactinoal infrastructure and to rely on the trusted, taken for granted background features that streamline human interaction
A term sometimes for the excesses of these effects is “AI psychosis”, but this is a form of victim-blaming that is best avoided: it uses mental health stigma to locate the effect solely in the affected user while failing to highlight the contributions of the “AI” system and its design. From an interactional perspective, the term distributed delusions is more apt (Osler 2026). As Osler points out, it is specifically their “conversational and validating style of engagement” that makes these systems so compelling.
But the present magic of “AI” is performed on us without our input, without our knowledge, and without our consent. There’s another word for that kind of magic; binding, subjugating magic performed on you against your will is called a curse. (from Williams 2023:108)
fire Fire is a physical phenomenon. Flame is the part of of fire.
Hallucinating with AI: Distributed Delusions and “AI Psychosis” Lucy Osler
Many psychometric reports bear a disconcerting resemblance to what my colleague Donald G. Paterson calls "personality description after the manner of P. T. Barnum" (13). I suggest—and I am quite serious—that we adopt the phrase Barnum effect to stigmatize those pseudosuccessful clinical procedures in which personality descriptions from tests are made to fit the patient largely or wholly by virtue of their triviality; and in which any nontrivial, but perhaps erroneous, inferences are hidden in a context of assertions or denials which carry high confidence simply because of the population base rates, regardless of the test's
Finally, if, as we have said, the sequences under consideration have a flat syntax, then it may be necessary to engage with a broader conception of grammatical structure, either within the generative framework or from compatible alternative frameworks. The former would include consideration of grammars that are not context-free or context-sensitive (e.g., regular or subregular grammars), while the latter would include work that considers indication and depiction alongside the well-studied mode of description (Ferrara & Hodge 2018), and work from typology and grammaticalization that shows why ideophones come to occupy the grammatical positions they do (Güldemann 2008). In both cases, there is a truly minimalist goal to achieve: to account for the observed structure by importing just the appropriate amount of syntactic machinery.
foto met ernstig tegenlicht van de laagstaande zon laat een deel van Kladderadatsch zien tegenover een mensenmassa die mooi van achteren uitgelicht wordt
Figuur 1. Niets nieuws onder de zon: “AI” sluit aan in een lange rij van technologieën met revolutionaire pretenties, maar goed onderwijs staat of valt met de kennis en kunde van leerkrachten, verankerd in gedeelde waarden. (Illustratie: Ada Jušić & Eleonora Lima (KCL) via betterimagesofai.org, CC -BY 4.0)
Montessori schrijft dat haar materialen een emancipatorische werking kunnen hebben, dat in regulier onderwijs een hiërarchische relatie tussen leraar en kind zelfstandig leren in de weg zit. Als de leraar het kind teveel helpt kan het minder gaan vertrouwen op de eigen intelligentie. In jouw artikel Human creativity in the age of text generators schrijf je dat chatbots laten denken dat ze een denkvriend (thinking buddy) zijn. Kun je meer zeggen over deze interactie met een chatbot? ‘Maria Montessori begreep dat leerkrachten soms het leerproces in de weg kunnen zitten, maar legde ook de nadruk op het belang van samen observeren en experimenteren. Het grappige is dat de introductie van een chatbot ons eerder terugbrengt bij wat Montessori beschreef als “de oude methode, waarin ze het kind overspoelen met nutteloze en vaak incorrecte woorden” – mijn vertaling. Om het leerproces zoveel mogelijk te faciliteren, hebben we leerkrachten nodig die het kind écht zien, en materialen die ontdekking op eigen kracht mogelijk maken. Chatbots bieden geen van beide.’
screenshot of fern generated by a simple recursive algorithm
I fe me )mere ke k) lo nya so a kyer) I siwu am3. Mm)n ngb) ta lo bra I un diary (leye n3 siwu iyere 😅)am3. Tela du3 n3 si boa wo deka Ido ng3. ife mɛmɛre ke kɔ lonya sɔ a tsɛrɛ i siwu amɛ. mmɛ ngbɔ ta lo'bra i un diary (leiye nɛ siwu iyere 😅) amɛ. Tela duɛ nɛ si boawo deka ido ngɛ "It makes me happy to see you writing in Siwu. As for me I keep a diary (I don't know the Siwu name). I'd love for us to meet some time."
Screenshot of Grammarly homepage advertising Proofreading, "Humanizer agent", "AI detector agent"
gloomy empty space showing the shadow of the former kitchen
Thank you for your interest! ResearchGate tries to keep you on its site to increase traffic and maximise profits, even though it knows this paper is published open access in Glossa here: https://www.glossa-journal.org/articles/10.5334/gjgl.444/ (Rant, feel free to ignore: Why don’t I just redundantly share the file here as well? Because ResearchGate parasitizes open access journals and resources by scraping their content, then pretending that same content is available only through ResearchGate. If RG operated in the spirit of scientific exchange, it would link to other resources where openly available, instead of hoarding content and making the original content providers harder to find.)
Cover page of For robust research, center values, not technology, by Mark Dingemanse and Christine Cuskley
Screenshot of ISCA website with "Resources" menu, showing Schegloff Transcription Module, Schegloff Media Archive, Schegloff Publications Archive, and under the Member-only resources the Schegloff Lectures, Schegloff Course Data, and Jefferson Index. The latter is Jefferson's index to Sacks' lectures, so not a representation of Jefferson's contributions or publications.
Samira Ibnelkaïd showing a slide with three talks that we're missing out on today: "Unsettling signs: Indexical Disorder and the Diacritics of Raciolinguistic Life" by Jonathan Rosa; "Recovering the Interactional Production of Inequality and Marginalisation" by Phil Hutchinson, and "Practices for Activity Transitions in Late 19th Century Immigrant Letters" by Nordlund et al.
Cover of Calabria 2026 The Grammar of Collaborative Turns Temporality, Projection, Emergence and Clause Combination in Italian Talk-in-Interaction
This is a nation of pushers and addicts This is a time of great change But the habits are deeply ingrained And we don't have the strength to rip up the old fabric The illusion is heavy And it takes a maverick To stand in the crosshairs and say, I won't have it
If anything, today’s text generators are more similar to the generations of bumbling male scholars who were happy to take credit for the work of their brilliant female peers.8 Except that text generators don’t even have the redeeming feature (if we can call it that) of being capable of original thought. Yes, they can produce novel arrangements of words that can be invested with meaning by eager interpreters9 — but to call that thought is the same as to attribute foresight to horoscopes. Text generators spin authoritative-sounding prose but have no sense of authorship and sunder all links of attribution. (Bropropriation is when a man takes credit for an idea first expressed by someone else.10 In this sense, text generators are the ultimate bropropriators.)
About me Wikipedian since August 14, 2004. Writing to learn. (And learning to write. I am not a native speaker of English. Don't ever hesitate to point out problems in my style, vocabulary or grammar to me - I'd like to learn. Thanks in advance!)
Door al dit werk, kritisch én constructief, staat de Radboud Universiteit in Nederland en wereldwijd bekend als een plek waar een helder en optimistisch tegengeluid klinkt, waar de kritische vermogens van staf en studenten serieus genomen worden, en waar geen genoegen genomen wordt met de marketinghype en het onafwendbaarheidsframe van Big Tech. De universiteit heeft hiermee goud in handen. Het CvB zou er goed aan doen dit veel prominenter te laten doorschijnen in haar strategie.
3. Wat betreft onderzoek herken ik de kritisch-constructieve aard van veel van het onderzoek aan onze universiteit niet in dit plan. Twee belangrijke steekwoorden in het plan zijn AI en duurzaamheid. Juist op het kruisvlak van die twee blinkt onze universiteit uit.
2. In bedrijfsvoering moet "AI" de universiteit gaan helpen om "processen efficiënter en duurzamer te maken" (p. 33). De cruciale vraag om welke vorm van “AI” het zou moeten gaan blijft onbeantwoord (er is ook helemaal geen “AI” die kan helpen bij bv. roostering of financiën, processen die al vergaand geautomatiseerd zijn en juist een menselijke maat behoeven). Prof. Tamar Sharon bij iHub heeft laten zien hoe dit soort technosolutionisme vaak leidt, niet tot oplossingen, maar tot technologisch gedreven herdefinities waardoor de echte problemen uit het zicht verdwijnen.
1. Wat betreft onderwijs neemt het document onkritisch de frames over van de makers van AI en doet het geen recht aan aan de breedte en diversiteit van expertise die we in huis hebben op dit domein. Wie dit document leest krijgt de indruk dat AI vooral "grote kansen" biedt (p. 21) en dat de primaire bijdrage van onze instelling ligt in "technische innovaties" en "concrete AI-oplossingen" (p. 32): een vorm van technosolutionisme.
0. Algemene indruk. Het document doet vooral heel glad en corporate aan. Om heel eerlijk te zijn herken ik er eigenlijk geen universiteit in, en zeker niet de onze. Dit zou ook over een instelling voor praktisch onderwijs kunnen gaan, of wellicht een technische universiteit: de nadruk ligt uitsluitend op toegepast nut van “AI” en er is te weinig oog voor de kritische wetenschappelijke en maatschappelijke functies van de universiteit
Screenshot of comment on my blog 'Jesse Sheidlower says: January 11, 2008 at 9:25 pm As I wrote in LanguageHat’s comments: The OED is citing a printed card announcing two of the London Dialectical Society’s November meetings, mailed by Ellis to James Murray (they were friends), and subsequently deposited by Murray in the OED archives. We’ll be clarifying our bibliography to show that this is not a published item.
Zwart-witfoto van een menigte op straat in Nijmegen met op de voorgront Frank Koster (snare) en Servaes Kusters (dikke trom). Net achter de trom is de zanger Peter Freijsen te zien. Omschrijving uit het archief: Straat- en Strijdorkest Kladderadatsch tijdens de door hen georganiseerde eerste Straatmuziekdag.
In March 2026, an Anthropic employee released the source code of Claude Code, a wrapper around their large language model that is widely used to generate code in programming tasks. Its thousands of lines of Typescript code contained many hopeful prompts and in cantations to shape Claude’s behaviour. Here are some examples: “Report outcomes faithfully”; “Never characterize incomplete or broken work as done”; “Be careful not to introduce security vulnerabilities ” (prompts.ts in Anthropic 2026) . There is more than a passing resemblance here to the Azande witch -doctor apprentice who, while stirring the medicine, utters: “You medicine which I am cooking, mind you always speak the truth to me. Do not let anyone injure me with his witchcraft, but le t me recognize all witches. … Let me be expert at the witch -doctor’s craft so that people will give me many spears on account of my magic.” (Evans-Pritchard 1937: 93). In the case of Claude, the incantation s appeared insufficient: analysis of the codebase, which according to a company executive was “pretty much 100% written by Claude Code ”, revealed severe security vulnerabilities (Townsend 2026).
For robust research, center values, not technology Mark Dingemanse* and Christine Cuskley *Corresponding author. doi:10.1017/S0140525X26104816 Large language models are interesting, but linguistics and cognitive science should be cautious about centering any new technology as a magic bullet. Doing so reinforces the historically “narrow focus” of linguistics identified in the target article, rather than expanding our understanding of human language. We argue that centering values instead of technology is the best way to future-proof scholarly work and to keep finding out new things about language.
Poster van de Kladderadatsch Deltatour. Een aantal foto's van Kladderadatsch en vier concertlokaties: Goese Sas, Goes, Veere, Zierikzee.
A four-quadrant diagram discussing arguments and research programs related to language models. Top left: "The String Statistics Strawman" with a faulty argument that LMs cannot learn language and refutations. Top right: "The "As Good As It Gets" Assumption" with a faulty argument that LMs are useless for serious language science and refutations. Bottom left: "The Statistical Structure Steelman Research Program" with key research questions on processing, learning, structure, and real patterns. Bottom right: "The "Yes, And" Research Program" listing shortcomings of current LMs and corresponding research opportunities.
Generative AI summaries make ACM authors' work look mediocre and bland and endanger ACM DL's mission
“Gods speak, spirits speak, computers speak. Oracular ambiguity or statistical probability provides loopholes, and discrepancies are expunged by Faith.” — Ursula K. Le Guin, The Left Hand of Darkness, 1969
Magic 8ball with ChatGPT branding
of the most common tropes used both in scientific research as well as in practical guides and tutorials casts prompt engineering as a “dark art” that “requires extensive trial and error” (Câmara et al. 2025) . One guide proposes that “If Claude is a genie in a bottle, then prompts are your wishes —and we all know what happens with poorly worded wishes” (Dickey 2025). In online tutorials, people share prompting tips and techniques in ways that are reminiscent of the sharing of magic al formulae, and a recurring recommendation is to affirm and specify the capability of the language model .3 Again divination offers parallels: “there are traditional refrains, pieces of imagery, compliments to the oracle, ways of formulating a question, and so forth which occur in every consultation” (Evans-Pritchard 1937: 137).
And finally, because their output is stochastic and not well-defined nor well-scoped [19], they are fundamentally unlike calculators [20, 21]. A common rhetorical frame is that just as pocket calculators replaced slide rules, so Generative AI will replace… what exactly? Working out the ways this analogy misrepresents the realities and complexities of scholarly research and writing is an exercise left to the reader.
Rather than letting ourselves be distracted by machines of mediocrity that have been built to capture our preferences and sell them back to us, we should focus our efforts on better understandingand celebrating the astonishing creativity and flexibility of human sense-making.
While most of these issues apply across societal spheres, there is something especially pernicious about text generators in academia, where writing is not merely an output format but a means of thinking, crediting, arguing, and structuring thoughts. Hollowing out these skills carries foundational risks. A common argument for Generative AI is a promise of higher productivity [5]. Yet productivity does not equal insight, and when kept unchecked it may hinder innovation and creativity [6, 7]. We do not need more papers, faster; we rather need more thoughtful, deep work, also known as slow science [8–10].
It is hard to forgive Frazer for his complacency and undisguised contempt of primitive society. The last chapter of Taboo and the Perils of the Soul is entitled, ‘Our Debt to the Savage’. Possibly it was inserted in response to correspondents who pressed him to recognise the wisdom and philosophic depth of primitive cultures they knew. Frazer gives interesting extracts from these letters in footnotes, but he cannot adjust his own preconceived judgments to take them into account.
Surely, a user says, you can’t deny that this exchange I had was meaningful to me? There is indeed no denying that. This is why a truly critical AI literacy must peel back the layers of human sense-making and investigate its interactional foundations.
In March 2026, an Anthropic employee released the source code of Claude Code, a wrapper around their large language model that is widely used to generate code in programming tasks. Its thousands of lines of Typescript code contained many hopeful prompts and incantations to shape Claude’s behaviour. Here are some examples: “Report outcomes faithfully”; “Never characterize incomplete or broken work as done”; “Be careful not to introduce security vulnerabilities” (prompts.ts in Anthropic 2026). There is more than a passing resemblance here to the Azande witch-doctor apprentice who, while stirring the medicine, utters: “You medicine which I am cooking, mind you always speak the truth to me. Do not let anyone injure me with his witchcraft, but let me recognize all witches. … Let me be expert at the witch-doctor’s craft so that people will give me many spears on account of my magic.” (Evans-Pritchard 1937: 93). In the case of Claude, the incantations appeared insufficient: analysis of the codebase, which according to a company executive was “pretty much 100% written by Claude Code”, revealed severe security vulnerabilities (Townsend 2026).
Six-panel composite figure. Caption: Interactive artifacts always rely on people’s interpretive and interactional practices. Rowwise from top left to bottom right: A. Aegeus consults the oracle at Delphi (cup from Vulci, 440-430 BCE). B. Byzantine mosaic depicting the zodiac, from the floor of the 6th century CE Beth Alpha synagogue. C. One-sided sense-making in an experimental psychotherapy session, (McHugh 1968). D. Still from a BBC documentary showing a person interacting with ELIZA via a computer terminal, late 1960s. E. Researchers interacting with the PARC copier (Suchman 2007 [1987]). F. Screenshot of large language model chat interface, 2026.
Watercolor painting of 12 zodiac signs. English months on one side of each sign, and another script on the other.
An electronic machine can carry out mathematical calculations, remember historical facts, play chess and translate books from one language to another. It is able to solve mathematical problems more quickly than man and its memory is faultless. Is there any limit to progress, to its ability to create machines in the image and likeness of man? It seems that the answer is no. It is not impossible to imagine the machine of future ages and millennia. It will be able to listen to music and appreciate art; it will even be able to compose melodies, paint pictures and write poems. Is there a limit to its perfection? Can it be compared to man? Will it surpass him? Childhood memories . . . tears of happiness . . . the bitterness of parting .. . love of freedom .. . feelings of pity for a sick puppy... nervousness ... a mother’s tenderness ... thoughts of death ... sadness ... friendship ... love of the weak ... sudden hope ... a fortunate guess ... melancholy ... unreasoning joy ... sudden embarrassment .. . The machine will be able to recreate all of this! But the surface of the whole earth will be too small to accommodate this machine — this machine whose dimensions and weight will continually increase as it attempts to reproduce the peculiarities of mind and soul of an average, inconspicuous human being. Fascism annihilated tens of millions of people.
The title of this chapter invokes a vast intellectual panorama; yet instead of vistas, I will offer only a twisting trail. The trail begins with some quite surprising crosscultural and crosslinguistic data, which leads inevitably on into intellectual swamps and minefields-issues about how our "inner languages" converse with one another, exchanging spatial information.
An image showing 5 figures from Laura van de Braak's work, with the section labels from the preprint: Oversimplification, Excessiveness, Misfocusing, Obfuscation, Exacerbation. The figures are described and captioned at length in section §4 of https://zenodo.org/records/17152893 (pp. 30ff)
Picture of Ursula Franklin next to a microscope, in lab coat, with a quote from The Real World of Technology: "Anyone who has ever woven or knitted knows that one can change patterns, but only at particular points and only in a particular way so as not to destroy the fabric itself. When women writers speak about reweaving the web of life, they mean exactly this kind of pattern change."
"Two pencil drawings of a map in two versions. On the left two points, A and B, separated by a mountain range. On the right two points, A and B, still separated by a mountain range, but this time with a line running straight from A to B"
Screenshot of page with figure. Caption says: "Two figures showing a pencil drawn map with points A and B and a mountain range. On the left, the mountain range is in between points A and B. On the right, the part containing the mountain range has been cut out, showing the author’s desk through the hole cut in the map.“ My hand-written note besides figure in earlier draft saying 'I love this one the most! I think you have a candidate cover image for the thesis!'
Image of hands and a loom, with quote from Le Guin: "But the word [technology] is consistently misused to mean only the enormously complex and specialised technologies of the past few decades, supported by massive exploitation both of natural and human resources."
"We need to keep up the fight."
Screenshot of Futures of Language landing page at https://futuresoflanguage.org, with tagline "Human interaction is too important to be left into the hands of merchants of hype." Text continues: "We study artisanal and artificial ways of languaging to better understand language + technology, and to reimagine our linguistic futures." Features a tricolour image of strands representing the three lines of work in the project.
Dazzling new capabilities have a way of obscuring sociotechnological complications (Suchman 2019). Thousands of papers published in the period 2021 -2023 relied on nominally free access to OpenAI’s text -davinci-002 language model. In late 2023, OpenAI suddenly phased out this model, undercutting at one blow the reproducibility of all this work (Liesenfeld et al. 2023). One affected study purported to show that “large language models are better than linguists”; its authors had to note that after text-davinci -002 was withdrawn, “the GPT 3 model that we were able to use as a replacement … is less sophisticated at understanding complex tasks” (Ambridge and Blything 2024). The result is a scientific edifice built on shifting sands, where “findings” have best-by dates haphazardly set by commercial parties.
We can play this game all day: seek permission for an edge case, retreat to another case when trouble arises, appeal to amiability or tolerance. Two things to note when a conversation gets to this point. First, it’s a misrepresentation: reluctance to grant permission is portrayed as an intolerant move, a wish to take the other’s toys, when it is nothing like that. What do you need my permission for, anyway? Second, it’s a distraction: it moves the conversation away from values and towards a tech-first perspective.
"We must transform our attitude toward all productive work and toward the planet into expressions of homemaking, where we create and sustain the possibility of life. It may take yet another new word to express the single responsibility that unites the homemaker, male or female, with the men and women who mine and plant and create industries and work for effective forms of exchange and for a peaceful world. Such a new term might be ecopoiesis" (p. 137)
To move from the specific to the general: Let’s make a checklist to help in the discourse on public decision-making. Should one not ask of any public project or loan whether it: (1) promotes justice; (2) restores reciprocity; (3) confers divisible or indivisible benefits; (4) favours people over machines; (5) whether its strategy maximizes gain or minimizes disaster; (6) whether conservation is favoured over waste; and (7), whether the reversible is favoured over the irreversible?
Picture showing planet Earth as photographed by DSCOVR:EPIC 00:13 on Wednesday February 25th, over the South Pacific Ocean Picture mostly shows a large expanse of blue ocean with some cloud cover spiralling over it.