Abdellah Fourtassi

Happy to share our new paper: "Scaffolding Early Dialogue: A Unified Account of Response Contingency in Child–Caregiver Interaction" with @abhiagrawal.bsky.social and Benoît Favre. Question: How can infants participate in dialogue while their linguistic and cognitive abilities are still immature? 🧵

Excited to present our new paper next week at EMNLP 2026 - we report new findings about how caregiver feedback can support children’s grammar learning (link to the paper below) 🧵

LM Playschool workshop: I’ll give a talk synthesizing our team research on interaction and language learning, of which this paper is one example. BabyLM workshop: Mitja Nikolaus will give a talk presenting his PhD research done in collaboration with Laurent Prévot and myself.

EMNLP paper: arxiv.org/pdf/2609.08576 Philo Trans B paper: royalsocietypublishing.org/rstb/article...

arxiv.orgarxiv.org

Excited to present our new paper next week at EMNLP 2026 - we report new findings about how caregiver feedback can support children’s grammar learning (link to the paper below) 🧵

This can happen without requiring an explicit teaching agenda: the signal can arise simply from the fact that caregivers are spontaneously less likely to reuse structures that are ungrammatical and more likely to reuse the grammatical ones

If you’re attending EMNLP, there are several opportunities to discuss this research: Main Conference: Jing Liu will present our paper as a poster on Monday (Hall H, Session 6 — Poster Session D, 11:00 AM – 12:30 PM)

Excited to present our new paper next week at EMNLP 2026 - we report new findings about how caregiver feedback can support children’s grammar learning (link to the paper below) 🧵

Our main finding is that structural alignment (reusing the child’s syntactic structure, without necessarily repeating the words themselves) produces the strongest improvements in grammatical production—even stronger than the effect of clarification requests reported in our previous work.

The significance here is that this provides evidence for a plausible developmental mechanism: caregivers’ tendency to reuse children’s grammatical structures can reinforce the use of these structures.

Excited to present our new paper next week at EMNLP 2026 - we report new findings about how caregiver feedback can support children’s grammar learning (link to the paper below) 🧵

This builds on our previous work, published earlier this year in Philosophical Transactions B (link below), where we focused on clarification requests as feedback.

In this new paper, co-led by Jing Liu and Marianne Schweitzer, we extend the methodology to other forms of feedback we extracted from child-caregiver conversations, including structural alignment, semantic contingency, and affective feedback.

Excited to present our new paper next week at EMNLP 2026 - we report new findings about how caregiver feedback can support children’s grammar learning (link to the paper below) 🧵

@rhodricusack.bsky.social

We use a small version of GPT-2 as a controlled learner and apply techniques inspired by reinforcement learning from human feedback (RLHF) to test the effects of different feedback signals extracted from natural child–caregiver conversations.

Abdellah Fourtassi reposted

Dirk Gütlin

@gutlin.bsky.social

📣📣📣 Preprint Alert! 📣📣📣 Do predictive processes shape brain representations during learning? We investigated the effect of Predictive Coding by training identical recurrent networks with different optimization procedures and then comparing them to human EEG under the same visual learning task.

Abdellah Fourtassi reposted

Rhodri Cusack

10 Sept

We're hiring! 2 postdocs + 3 PhD students to join InfantNeuroAI at Trinity College Dublin: awake infant fMRI & OPM-MEG, plus computational models of how infants learn. Aims: understand infant visual cognition, infant neuroimaging methods, more energy-efficient learning 🧵 www.cusacklab.org/vacancies

Abdellah Fourtassi reposted

Riccardo Fusaroli

@fusaroli.eurosky.social

10 Sept

How does the architecture of a city shape the way we think, talk, and draw space? And how do spatial conventions travel across cultural generations? New paper in Open Mind, spearheaded by Z Alinam (w J Nölle, C Vesper, K Tylén & me): doi.org/10.1162/OPMI.a.380 🧵 1/

doi.orgWhen the City Speaks: How Urban Form Shapes the Cultural Transmission of Spatial ConceptualizationAbstract. How do urban environments shape the way people conceptualize and communicate space? Using a cultural transmission experiment in virtual reality, we asked participants to navigate one of two ...

Abdellah Fourtassi reposted

Peter Kok

@peterkok.bsky.social

9 July

The hippocampus sits at the interface of perception & memory, allowing us to link past, present, & future through its role in generating predictions. Excited to share this theme issue in @royalsociety.org's Philosophical Transactions B, co-edited with @barense.bsky.social & @mariamaly.bsky.social

royalsocietypublishing.orgVolume 381 Issue 1954 | Philosophical Transactions of the Royal Society B | The Royal SocietyInfluential themed journal issues across the life sciences.

Abdellah Fourtassi reposted

Marianne de Heer Kloots

@mdhk.net

3 July

Now out in BBS, as commentary on @futrell.bsky.social & @kmahowald.bsky.social's "How linguistics learned to stop worrying and love the language models"! Humans learn much of spoken language structure from speech (not text), & we can study models that do the same. www.cambridge.org/core/journal...

Abdellah Fourtassi reposted

Grzegorz Chrupała

@grzegorz.chrupala.me

5 Sept

Accepted to EMNLP 2026: "Beyond Decodability: Reconstructing Language Model Representations with an Encoding Probe". Probing asks what you can decode from a model's activations. We reverse it: reconstruct activations from interpretable features, then ablate to rank them. arxiv.org/abs/2605.00607

arxiv.orgBeyond Decodability: Reconstructing Language Model Representations with an Encoding ProbeProbing is widely used to study which features can be decoded from language model representations. However, the common decoding probe approach has two limitations that we aim to solve with our new enc...

onlinelibrary.wiley.comScaffolding Early Dialogue: A Unified Account of Response Contingency in Child–Caregiver InteractionEarly verbal dialogue emerges during the second year of life, despite children's still-developing linguistic, pragmatic, and cognitive abilities. This raises a puzzle: how can young children produce ...

Link to paper (open access): onlinelibrary.wiley.com/doi/10.1111/...

I am particularly proud of this work. Beyond the scientific contribution, it is the culmination of several years of research aimed at bringing together theoretical and practical insights from computational dialogue systems to inform theories of child language development in interaction

Mechanism 3: response modeling. Children are more contingent for communicative intents that caregivers typically express with short utterances.

All three mechanisms remain predictive in a joint model controlling for age, child utterance length, and intent frequency. Together, the results support a unified account in which caregivers reduce the processing demands of early dialogue.

Mechanism 1: lexical specificity. Children respond more contingently when caregivers use linguistic forms that are more diagnostic of their communicative intent. The pattern is consistent across corpora.

Mechanism 2: interactive routines. Contingency rises sharply when caregivers initiate predictable sequences—such as questions or requests—and children produce the expected response type.

Producing a contingent response requires several coordinated steps: decoding the speaker’s intent, selecting an appropriate next intent, and realizing that intent linguistically—all under the time pressure of turn-taking. We ask whether caregivers scaffold each of these steps.

@lucaswbenjamin.bsky.social

We test this at scale: 40 CHILDES corpora, 609 children, and 2,577 conversations, focusing on children around 20–32 months.

Abdellah Fourtassi reposted

Lucas Benjamin

24 Aug

Our new paper is now out in @pnas.org! How does the (baby) brain extract regularities from sequences from what-follows-what up to large-scale network structure? What if it all came from a single mechanism? www.pnas.org/doi/abs/10.1...

Abdellah Fourtassi reposted

Neil Burgess

@neilburgess10.bsky.social

8 Aug

Great to see Ellie Spens’ model out! Interplay btwn episodic (hippo) & semantic (neocort’l) systems, in encoding, consolidatn, recall & problem solving with sequential info, as compressive retrieval-augmented generation. Inc gist-based distortion in stories & inference www.nature.com/articles/s41...

nature.comHippocampo-neocortical interaction as compressive retrieval-augmented generation - Nature CommunicationsHere the authors model interactions between episodic and semantic memory as ‘retrieval-augmented generation’, explaining consolidation of structure from sequential experience, the reconstruction and d...

Abdellah Fourtassi reposted

Marianne de Heer Kloots

@mdhk.net

30 June

Let’s study learning trajectories in self-supervised speech models! 🔊 Do they reflect the hierarchical organization of spoken language? We have analyzed a lot of training checkpoints to find out 🌠 Preprint: arxiv.org/abs/2604.02043 ⬇️

Abdellah Fourtassi reposted

Giulia Rambelli

@grambelli.bsky.social

31 July

🎉 Happy to have presented my postdoctoral work in collab with Laurent Prévot & @fourtassi.bsky.social at #ESLP2026 ! 🧠 Can distributional input alone support the emergence of abstract relational knowledge? We explore this through computational models of child-directed speech. #NLPproc #cogsci

Abdellah Fourtassi reposted

Greg Woodin

@gregwoodin.bsky.social

29 July

Had fun presenting our work (@gretagandolfi.bsky.social, @gabriellavigliocco.bsky.social, @fourtassi.bsky.social, Yan Gu) on multimodal coordination in early word learning w 3–4-year-old children and their caregivers today at Embodied and Situated Language Processing at Lancaster University. (1/4) 👇

Abdellah Fourtassi reposted

CoNLL 2026

@conll-conf.bsky.social

24 June

We are excited to announce the second keynote of CoNLL 2026! Adele Goldberg (Princeton University) “Compositionality and creativity in natural language and LLMs” 📅 Saturday, July 4, 2026 🕘 09:00 📍 Room Harbor B #CoNLL2026 #NLP #Compositionality

Abdellah Fourtassi reposted

Greg Woodin

@gregwoodin.bsky.social

5 May

My current position will be coming to a close at the end of 2026, so I’m beginning to explore next steps. I’m a postdoctoral researcher at UCL working at the intersection of language, cognition, and multimodal communication. 1/4 👇

Abdellah Fourtassi reposted

Greg Woodin

@gregwoodin.bsky.social

23 Apr

Does verbal coordination help 3–4 year old children learn new words? In our new preprint (w/ @gabriellavigliocco.bsky.social, @gretagandolfi.bsky.social, @fourtassi.bsky.social, Yan Gu), we explored 20 predictors of word learning. One combination stood out. 👇 (1/4) papers.ssrn.com/sol3/papers....

huds.uni-goettingen.dehuds.uni-goettingen.de

21 Apr

Are you an advanced PhD student looking for a reviewing experience? I’m looking for one emergency reviewer for one paper in ACL Rolling Review in the track "Linguistic Theories, Cognitive Modeling, and Psycholinguistics" --due before next Tuesday April 28th.

Abdellah Fourtassi reposted

Nivi Mani

@nivimani.bsky.social

3 Apr

Call for a PhD position in Cognitively inspired natural language processing with Lisa Beinborn and me. Part of the new SPP "Robust Assessment & Safe Applicability of Language Modelling: Foundations for a New Field of Language Science & Technology” (LaSTing). huds.uni-goettingen.de/assets/Call_...

Abdellah Fourtassi reposted

Caroline Rowland

@carorowland.bsky.social

31 Mar

Applications open for the School on Analytical Connectionism 📅 August 17-28, 2026 🗺️ Chalmers University of Technology Gothenburg 📚 Topical focus: language acquisition ... with me, Michael Biehl, Paul Smolensky & many other incredible researchers 😊 www.analytical-connectionism.net/school/2026/

analytical-connectionism.net2026 School on Analytical ConnectionismA 2-week summer course hosted at Chalmers University of Technology on analytical approaches to language acquisition and higher-level cognition.

Abdellah Fourtassi reposted

Marvin Lavechin

@marvinlavechin.bsky.social

9 Mar

What if you could automatically transcribe children's speech sounds from their first babbles to full sentences? Screening for speech delays. Comparing how kids learn to talk across languages. Following how sounds evolve month by month. We're building toward this with BabAR🧵 (sound on 🔊)

8 Mar

Open PhD/Postdoc position (start: Oct 2026). Topic: AI/LLMs and child language/communicative/cognitive development. The exact project will be shaped with the candidate. Join our team @univ-amu.fr at the intersection of computer and cognitive science (& right next to the Calanques!). Send me your CV!

6 Mar

Deadline approaching (March 20), consider submitting! The process is flexible to accommodate different situations: - Hybrid presentation mode - Direct submission or via ARR - Non-archival option: present work that is (or will be) published elsewhere - Short (4 pages) or long (8 pages) papers

Abdellah Fourtassi@fourtassi.bsky.social

Thrilled to announce the 1st Workshop on Computational Developmental Linguistics (CDL) at ACL 2026 🎉 A new venue at the intersection of development linguistics × modern NLP, spearheaded by @fredashi.bsky.social @marstin.bsky.social, and and outstanding team of colleagues! A thread 🧵

Abdellah Fourtassi reposted

micha heilbron

@mheilbron.bsky.social

5 Mar

📢 PhD position in the NeuroAI of Language Why can LLMs predict brain activity so well? We're hiring a PhD student to find out -- AI interpretability meets neuroimaging Deadline March 20 Please RT 🙏 👇 mpi.nl/career-education/vacancies/vacancy/fully-funded-4-year-phd-position-neuroai-language