Mark Dingemanse
LightNews LightNews
banner
dingemansemark.bsky.social
Mark Dingemanse
@dingemansemark.bsky.social
2.6K followers 420 following 4.1K posts
Language, interaction, tech • Here with doubts about the next monetizable monopoly... • papers https://markdingemanse.net • blog https://ideophone.org • fedi https://scholar.social/@dingemansemark/ • POSSE: Publish on Own Site, Syndicate Everywhere
Posts Replies Media Videos
dingemansemark.bsky.social
Iconicity ratings (every dot represents a single ideophone). A: Distribution of mean iconicity ratings. B: Distribution of ratings by cumulative iconicity, showing a stepwise increase in iconicity ratings and a narrowing distribution of variance as cumulative iconicity goes up.
November 8, 2025 at 2:24 PM
dingemansemark.bsky.social
One year ago, with Stella Punselie and Bonnie McLean: The Anatomy of Iconicity, in Open Mind doi.org/10.1162/opmi...

In which we combine iconicity ratings, experimental evidence, and structure mapping to identify & explain crosslinguistic patterns of iconic words

#iconicity #cognitivescience
Overview of the triangulation method. Structure mapping grounds the notion of iconicity, ratings capture subjective form-meaning fit, and guessability provides a baseline of guessing accuracy. For instance, Korean tuˈgɯndugɯn ‘heartbeat’ features at least three levels of iconic form-meaning correspondences (A). This predicts it should be rated as highly iconic (B) and should be more guessable than words with fewer detected correspondences (C). Consistent with this, it was rated 4.8 for subjective iconicity and guessed at 80% accuracy. Arrows indicate how measures predict (black) and inform (grey) each other.
November 8, 2025 at 2:24 PM
dingemansemark.bsky.social
Ah! Blijkt dat ik Mulatos al kende, maar de andere albums nog niet, en ik was 'm de laatste jaren duidelijk vergeten www.last.fm/user/strange...
Last.fm listening history showing my plays of Omar Sosa, mostly in 2015-2019
October 26, 2025 at 11:48 AM
dingemansemark.bsky.social
not sure about insightful but I like how social interaction offers a natural laboratory to study convergent cultural evolution of linguistic items, as in this study of ours combining comparative evidence and a simple computational model pure.mpg.de/rest/items/i... (pdf)
CONVERGENT CULTURAL EVOLUTION OF CONTINUERS (mmhm) MARK DINGEMANSE*1,2, ANDREAS LIESENFELD1, and MARIEKE WOENSDREGT2 *Corresponding Author: mark.dingemanse@ru.nl 1Center for Language Studies, Radboud University 2Max Planck Institute for Psycholinguistics Continuers —words like mm, mmhm, uhum and the like— are among the most frequent types of responses in conversation. They play a key role in joint action coordination by showing positive evidence of understanding and scaffolding narrative delivery. Here we investigate the hypoth- esis that their functional importance along with their conversational ecology places selective pressures on their form and may lead to cross-linguistic similarities through convergent cultural evolution. We compare continuer tokens in linguistically diverse conversational corpora and find languages make available highly similar forms. We then approach the causal mechanism of convergent cultural evolution using exemplar modelling, simulating the process by which a combination of effort minimization and functional specialization may push continuers to a particular region of phonological possibility space. By combining comparative linguistics and computational modelling we shed new light on the question of how language structure is shaped by and for social interaction. 1. Int
October 21, 2025 at 4:42 PM
dingemansemark.bsky.social
The latest issue of @europeanreview.bsky.social is so gorgeous and interesting
Spread from ERB showing botanical illustration of Cornelian cherry as part of a story by Kapka Kassabova
October 11, 2025 at 10:14 AM
dingemansemark.bsky.social
Lucy Suchman! Plenty of relevant work in the past decade, here is just one example www.tandfonline.com/doi/abs/10.1...
Building upon generative intersections of critical security studies and science and technology studies (STS), I argue that the promotion of automated data analysis under the sign of artificial intelligence can only serve to exacerbate military operations that are at once discriminatory and indiscriminate in their targeting, while remaining politically and legally unaccountable.
September 10, 2025 at 1:02 PM
dingemansemark.bsky.social
My rant from a few years back about the cursed little histograms isn't about this but of course it is a side-effect. ideophone.org/a-rant-about... Elsevier benefits from the SEO juice of uni domains and gives nothing back, enshittifying pages with fluffy management-bait like SDGs & Fingerprint
September 5, 2025 at 9:57 AM
dingemansemark.bsky.social
We have updated our post. Will you update your marketing materials to tone down the open-washing? osai-index.eu/news/lumo-pr... #OpenWashing
Update September 1 Proton's Lumo account on BlueSky has responded to this post saying "Lumo isn't a model; it uses open models, including the model listed as the best for openness in this piece. You can find the models Lumo uses in our privacy policy". That page repeats the open source claim: Lumo’s code is open source, meaning anyone can see it’s secure and does what it claims to. We’re constantly improving Lumo with the latest models that give the best user experience. The only open source code we have found is for the Lumo mobile and web apps. Proton calling the Lumo AI assistant open source based on that is a bit like Microsoft calling Windows open source just because there's a github repository for Windows Terminal.
September 1, 2025 at 8:23 PM
dingemansemark.bsky.social
Not unaware. @proton.me knows privacy and FOSS concerns are key to its userbase; their users have pointed out the wrong claims from day 1, yet Proton has doubled down, saying (astonishingly) the statement merely reflects a "long-term intention" 😂 discuss.privacyguides.net/t/proton-mar...
Naturally, I wanted to check out the code. When I couldn’t find it, I reached out to support to ask for a link. After some back and forth, this was the official response I received: "The statement on our website reflects our long-term intention and the values we stand for, not necessarily the instantaneous state upon launch.”
August 31, 2025 at 9:09 PM
dingemansemark.bsky.social
Att this point, @proton.me has not provided any documentation of the Lumo system despite touting its 'open source' status in three different places. It could be a finetune of Mistral or OLMo; a big tech LLM API; or multiple in a trenchcoat. Point is, we would know if it were truly open source 🫠
Screenshot of Lumo promotion website with claim highlighted: "Opens source code to the public". Lumo in the OS AI ibdex: nothing open about it
August 30, 2025 at 8:19 AM
dingemansemark.bsky.social
Reminder! Less than a week left to apply for PhD and postdoc positions in Futures of Language — 🔁 for folks interested in the intersection of language, social interaction, and technology

Vacancies: markdingemanse.net/futures/news...
FAQ: markdingemanse.net/futures/news...

#emca #linguistics #sts
View of Radboud Campus in Nijmegen, brutalist Erasmus building hosting the Faculty of Arts & the Futures of Language project in the foreground
August 20, 2025 at 6:37 PM
dingemansemark.bsky.social
bare minimum for holding a conversation

#emca
(1) Speaker-change recurs, or at least occurs (cf. $4.1, below). (2) Overwhelmingly, one party talks at a time (cf. $4.2). (3) Occurrences of more than one speaker at a time are common, but brief (cf. $4.3). (4) Transitions (from one turn to a next) with no gap and no overlap are common. Together with transitions characterized by slight gap or slight overlap, they make up the vast majority of transitions (cf. $4.4). (5) Turn order is not fixed, but varies (cf. $4.5). (6) Turn size is not fixed, but varies (cf. 44.6). (7) Length of conversation is not specified in advance (cf. $4.7). (8) What parties say is not specified in advance (cf. $4.8). (9) Relative distribution of turns is not specified in advance (cf. $4.9). (10) Number of parties can vary (cf. $4.10). (11) Talk can be continuous or discontinuous (cf. 44.11). (12) Turn-allocation techniques are obviously used. A current speaker may select a next speaker (as when he addresses a question to another party); or parties may self-select in starting to talk (cf. $4.12). (13) Various 'turn-constructional units' are employed; e.g., turns can be projectedly 'one word long', or they can be sentential in length (cf. $4.13). (14) Repair mechanisms exist for dealing with turn-taking errors and violations; e.g., if two parties find themselves talking at the same time, one of them will stop prematurely, thus repairing the trouble (cf. 44.14). Sacks et al. 1974
August 18, 2025 at 1:42 PM
dingemansemark.bsky.social
Great post about a fascinating phenomenon. We touched on this in a 2023 piece; it seems people have not really caught on in the interim, a testimony to the enduring effectiveness of RLHF in producing strong Eliza effects doi.org/10.1016/j.ti...
Whether following predesigned dialog flows or generated by next token prediction, the output of conversational interfaces is fundamentally devoid of accountability: it is text produced without a commitment to truth or joint action [72]. By tailoring textual output to be maximally in line with what human raters prefer, techniques like reinforcement learning with human feedback obscure this lack of accountability but do not fundamentally change it: a text generator may ‘apologise’ when prompted for clarification or correction but will indifferently regurgitate the same misunderstanding or a new one on the next occasion
August 8, 2025 at 7:46 PM
dingemansemark.bsky.social
Fellow EU OSAI Index researcher Nityaa Kalra has added ten new music models to our index along with a blog post about the MusGO project osai-index.eu/news/musgo-f...

All audio models here: osai-index.eu/the-index?ty...
Full index with >150 models: osai-index.eu/the-index #GenAI #OpenSource
Ranking of >10 audio generative AI models by degree of openness
July 25, 2025 at 6:59 AM
dingemansemark.bsky.social
Hey EditorialManager. I just want to file my review. You want me to fill out my profile in excruciating detail before I even get there.

#EditorialManager continues to excel in the mud run approach to user interface design. Finding reviewers is hard these days. Why not make it harder?
Screenshot of cursed Editorial Manager dialog window "Edit Personal Keywords". The three keywords are "DON'T" "FORCE" "ME"
July 22, 2025 at 11:41 AM
dingemansemark.bsky.social
Speaking of Le Guin and serendipity, I picked up The Left Hand of Darkness the other day just as I was thinking about 'AI', oracles and divination and my eye fell on this gem
Ursula Le Guin in The Left Hand of Darkness (1969): Legends of prediction are common throughout the whole Household of Man. God speaks, spirits speak, computers speak. Oracular ambiguity or statistical probability provides loopholes, and discrepancies are expunged by Faith.
July 16, 2025 at 7:51 PM
dingemansemark.bsky.social
Nimma! Morgen na het vuurwerk 🎇 speelt @kladderadatsch.bsky.social de sterren van de hemel bij podium Kaaij Hoog. Kom langs en waag een dansje!

zo 13 jul 23:00 - 00:30

Onderdeel van de #vierdaagsefeesten  #Nijmegen  #4daagse #4df www.vierdaagsefeesten.nl/programma/kl...
De aanstekelijke muziek die Kladderadatsch maakt is afkomstig uit vele culturen en volken: de Balkan met zijn melancholische Slavische klanken, de opzwepende muziek uit Zuid- en Midden-Amerika, de weemoedige Jiddische Klezmer, vrolijke Afrikaanse soul.
July 12, 2025 at 6:32 PM
dingemansemark.bsky.social
Sho presents his work validating dynamic time warping to a rapt #ISGS2025 audience. Lots of engagement afterwards #ProudSupervisor
Sho Akamine presenting on dynamic time warping on a large stage at ISGS 2025 in Nijmegen Sho answering questions
July 11, 2025 at 2:10 PM
dingemansemark.bsky.social
This reminds me of an amazing case in my Siwu corpus which I described & analysed in my 2011 thesis — a participant breaking into song in the local funeral dirge style. Ch12 in the thesis, audio sample at ideophone.org/thesis/part-...

PDF of (hefty) thesis: pure.mpg.de/pubman/item/...
personal telling about Kɔdzo (line 19), occasioned by two turns by C. Indeed, one of the striking things about D’s sung utterances is the fact that they are so well integrated into the sequence as a whole. D slips into the style as easily as she gets out again; there are no clear boundaries or changes of body position that might indicate a shift in genre, and C treats D’s singing not as an interruption of the sequence but as doing agreement and affiliation (and inviting the same in return). Extract 12.4 D’s lament (lines 8-15 from Extract 12.3 above) iɔ̀loo, iɔ̀lo kananaa iɔ̀loo, iɔ̀loo, kàgbàmìkù ɔɔ̀loo, iɔ̀loo kàgbàmìkù ɔɔ̀loo, ka-ɔ̀-loo, kàgbàmìkù ɔɔ̀loo, kàgbàmì-... It’s become still, still kananaa Still, still, the neighbourhood is still, still The neighbourhood is still, it is still, the neighbourhood is still, the neighb-... What can we learn from this lapse into sìnɔ (dirge) style midway in an affiliative sequence? The case itself would warrant a separate study as it raises fundamental issues like the boundaries of speech and song, composition in performance and intertextuality (e.g. Feld 1999:241ff. and references therein), but this is not the place for that. On purely local considerations, the lapse into dirge style is treated by C as an appropriate move towards the co-constructed goal of holding the neighbourhood responsible. Partial transcript showing two lines of speech and one more songlike
July 10, 2025 at 5:48 PM
dingemansemark.bsky.social
Interesting!

Puzzling choice that limits scientific uses: This is prompted, and they've broken up all 1h dyadic recordings into chunks of 2-10 min and left out what they call 'meta' moments where participants study instructions, ask clarification, etc

not sure why they call it Seamless
2.2.1 Conversation Based Prompt Creation We attempt to anchor each 2-10 min long dyadic interaction in the IPC via the use of prompts. Prompts are presented separately to each participant in an interaction prior to the start of the interaction. The majority of meta time in the dataset (which is not included in this release) is the short intervals between active interactions, during which participants are studying their prompts and asking clarifying questions of the moderators (refer to Section 2.1 for more detail on active and meta time).
July 10, 2025 at 7:11 AM
dingemansemark.bsky.social
Whether you are pondering applying for one of the above positions or just looking for some good reads, I think you will enjoy these summer reads markdingemanse.net/futures/news...

#SummerReads #linguistics #emca #sts #hci #FuturesOfLanguage
Covers of 4 books: Bridle, James. 2022. Ways of Being: Animals, Plants, Machines: The Search for a Planetary Intelligence. New York: Farrar, Straus and Giroux. Franklin, Ursula M. 1999. The Real World of Technology. Revised edition. CBC Massey Lectures. Toronto: Anansi. Illich, Ivan. 1973. Tools for Conviviality. Open Forum. London: Calder and Boyars. Le Guin, Ursula K. 2019. The Carrier Bag Theory of Fiction. London: Ignota.
July 9, 2025 at 2:50 PM
dingemansemark.bsky.social
I often use this slide comparing Utrecht, NL, in 1976 vs 2022. People made a deliberate choice against a car-centred city and for more convivial modes of living and moving about.

Industries & profiteers love inevitabilism about technology; it is up to us to organise and resist, and it can be done
Slide comparing two pictures of the same street in Utrecht: in 1976 there were cars parked along the sides and two car lanes in the centre, not a human, bike or tree in sight; in 2022, there are trees, benches, parking spots for bikes, terraces, and plenty of people moving about. Not a car in sight. Atop the slide is a quote from Ivan Illich: “The present world is divided into those who do not have enough and those who have more than enough, those who are pushed off the road by cars and those who drive them.” — from Tools for Conviviality
July 7, 2025 at 8:45 PM
dingemansemark.bsky.social
We're hiring! Join us to work at the intersection of social interaction and language technology. Postdoc and PhD positions in my Futures of Language research group, based at Radboud University in Nijmegen, NL

Read more: markdingemanse.net/futures/news...

#linguistics #interaction #sts #emca #hci
Aerial photograph of Radboud University campus with the Erasmus tower in the foreground: tallest building in Nijmegen & home to the Futures of Language project. Source: RU
July 7, 2025 at 11:44 AM
dingemansemark.bsky.social
Maar die kans hebben werknemers al ruimschoots, niet? Het is momenteel moeilijker om er aan te ontkomen — of is jouw, ervaring anders? Welke 'werkprocessen' aan de universiteit heb jij voor ogen?
Het begint al met de framing. Werknemers moeten “de kans krijgen om gebruik te maken van nieuwe technieken”. Newsflash: die hebben ze al. Microsoft schotelt ongevraagd alle MS365 gebruikers CoPilot voor (en verhoogt de prijzen). Er is geen universiteit waar deze technologie niet enthousiast binnengehaald is. Er is geen vakgroep waar niet geroepen is, al dan niet met een knipoog, “ik vraag het wel even aan ChatGPT”. Er is geen teaching and learning centre waar men niet onder de indruk is van de lesplannen en werkvormen die “AI” uit de mouw schudt.
July 3, 2025 at 7:28 AM
dingemansemark.bsky.social
Het bizarre is dat de bonden met deze AI-prietpraat de bal voor open doel neerleggen. Meer AI = minder werk, de bonden zeggen het zelf. Moet ik de volgende stap nog uittekenen? ideophone.org/waarom-ik-ni...
July 3, 2025 at 7:18 AM