iiinova — where your digital self keeps evolving

Reposted by Marco Ciapparelli

Fabio Marson @fabiomarson.bsky.social · 24/09/2026

Our big dataset on pseudowords and nonwords processing is finally out! 🎉🥲 It's been a long adventure but definitely worthed it. Thanks to my amazing colleagues for the incredible job done!

021

Reposted by Marco Ciapparelli

Nicole Rust @nicolecrust.bsky.social · 18/07/2026

Wonderful to see this! For (controversial?) context. There’s long been an argument that what brains & ANNs are doing cannot be fathomed beyond the meta like (eg) architecture, learning rules and such. And there’s a counter-idea: ”let’s try?”. When we stumbled on the correlates of memorability /1

2296

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

11/ Data (including our synthetic images), code and our preregistered plans are available at: osf.io/dp2u6/overview

osf.io

OSF

000

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

10/ We test 3 preregistered hypotheses, finding that the inferred appearance of compounds aligns with human coarse-grained preferences and judgment speed, but not with fine-grained preferences (image typicality). Comparisons of language models has implications for theories of conceptual combination.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

9/ Having established the above, we turn to zero-shot learning. We devise a set of novel compounds, each admitting multiple interpretations, each depicted by different synthetic images. Participants read the words, imagine their appearance and select images that best match their intuition.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

8/ First, using familiar concepts unknown to our vision models, we confirm that image features can be approximated by words’ distributional profiles. Then, we turn to human behavior. For *familiar* concepts, language predictions recover image typicality judgements (the "best depiction" of a word).

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

7/ We use computational models to characterize conceptual systems, their alignment, and combination: word embeddings capture the distributional profiles of words, a linear model combines them, while image features are extracted from CNNs, and linear regression maps concepts from language to vision.

110

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

6/ Our proposed solution is to rely on compounding. Novel compound words (e.g., “tiger collar”) are interpretable novel words: words whose usage profile can be induced on-the-fly by combination, and then used to make cross-modal, zero-shot predictions about appearance.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

5/ ...but creating naturalistic novel concepts is challenging: novel concepts not only have unknown appearances, but words that describe them, by definition, lack a history of usage.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

4/ It's not enough for concepts to be novel; they should also be similar to familiar concepts (“naturalistic”) to be situated in, and predictable from, the available knowledge base...

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

3/ In models, zero-shot learning is conceptually easy to test: hold out concepts from training – words or images – then use them as “novel” test cases. With humans this is harder, as one cannot precisely know or manipulate the learning history of a person. Thus, novel concepts must be engineered.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

2/ Words that are used similarly tend to refer to similarly looking concepts. This second-order similarity suggests that if we know how a concept is talked about, we might infer its appearance in a zero-shot fashion.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 30/06/2026

🚨 New preprint! 🚨 1/ Can word usage help us infer what novel concepts look like? New work on zero-shot learning in humans with Marco Marelli & Sharon Thompson-Schill! 🧵 osf.io/preprints/ps...

170

Reposted by Marco Ciapparelli

Neuroskeptic @neuroskeptic.bsky.social · 03/05/2026

Made-up non-words produce semantic brain activity "comparable to that of existing words, provided that novel words contain a concatenation of identifiable morphemes." www.biorxiv.org/content/10.6... This is a bluty result, but not too jarfish if you doot ahugeapun, no?

biorxiv.org

0284

Reposted by Marco Ciapparelli

ERCbravenewword @ercbravenewword.bsky.social · 20/04/2026

Truly enjoyed hosting @bottinilab.bsky.social and his team for our joint lab meeting. A fantastic three-day exchange where everyone shared their research and explored new ideas together. Here is Roberto Bottini’s seminar on cognitive maps: www.youtube.com/watch?v=j3iv...

youtube.com

Mapping the Mind: How Cognitive Maps Guide Mental Action | Roberto Bottini | BraveNewWord | Unimib

YouTube video by Mbs Vector Space Lab

041

Reposted by Marco Ciapparelli

PLOS Biology @plosbiology.org · 15/04/2026

How do we navigate our #memory maps of spatial & non-spatial experience? This study shows that the #EntorhinalCortex not only encodes the structure of experience, but also actions in conceptual space, and these representations are reflected in #gaze behavior @plosbiology.org 🧪 plos.io/4czCPHk

Experimental design. Top left: Representations of (conceptual) space may benefit from the integration of action affordances, e.g., in a game of chess, where information about how pieces can move is necessary to play the game. The authors tested this idea using numerical operations, whereby participants moved between numbers following the rules of an underlying graph structure. Top right: Participants learned how numbers may ‘transform’ into other numbers using an exploration task, whereby they were presented with an individual number and were offered a free choice of two possible successors. When a successor was chosen, the next trial offered the successors for the chosen number. In interleaved test blocks, participants indicated which were the ‘correct’ successors for each number. In the eyetracker and fMRI scanner, individual numbers were presented in a pseudorandom order. Bottom: Unbeknownst to participants, number-action bindings formed part of a repeating sequence arranged in graph ‘modules’ of four numbers. This can be seen as a set of four states with the same action affordances, projected on a number line. In the example presented, numbers 21 and 26 correspond to state 1 (‘S1’; and therefore also 31, 36, etc.), but this was assigned differently to each participant. As the modules repeated every five numbers, the structure allowed easy inference of action possibilities for any new numbers.

0101

Reposted by Marco Ciapparelli

ERCbravenewword @ercbravenewword.bsky.social · 01/04/2026

How do cognitive maps guide our actions in both spatial and conceptual worlds? Seminar by Roberto Bottini (CIMeC, UniTrento) on hippocampal–entorhinal systems and conceptual navigation. 🗓️ Wed, April 15 | 3:30 PM 📍 Bicocca (U6 - Sala Lauree) 🌐 Online: meet.google.com/yne-wucy-uju

021

Reposted by Marco Ciapparelli

tbiba.bsky.social @tbiba.bsky.social · 02/03/2026

I am excited to share my first paper, showing that episodic memory formation is theta rhythmic, is now published in Nature Human Behavior! Check it out here: rdcu.be/e6pzS. Thanks to my PI, Katherine Duncan, and to my collaborators for their support on this journey! Stay tuned for iEEG follow up 🧠

rdcu.be

Episodic memory encoding fluctuates at a theta rhythm of 3–10 Hz

Nature Human Behaviour - Biba et al. show that episodic memory encoding fluctuates at a theta rhythm of 3–10 Hz.

312346

Reposted by Marco Ciapparelli

Gary Lupyan @glupyan.bsky.social · 19/01/2026

arxiv.org/abs/2601.11432 I want to share an astonishing result. LLMs can "translate" Jabberwocky' texts like 'He dwushed a ghanc zawk” & even and even 'In the BLANK BLANK, BLANK BLANK has BLANK over any BLANK BLANK’s BLANK' This has profound consequence for thinking about.. 1/2

arxiv.org

1513734

Reposted by Marco Ciapparelli

Tanja C Roembke @troembke.bsky.social · 17/12/2025

@matildellen.bsky.social bravely decided to do a registered report with @pci-regreports.bsky.social as the first manuscript of her PhD -- 2+ years later, it's out! It's on how bilingualism affects statistical word learning, and the results were quite unexpected. Read more:

royalsocietypublishing.org

The influence of bilingualism on statistical word learning: a registered report

Abstract. While statistical word learning has been the focus of many studies on monolinguals, it has received little attention in bilinguals. The results o

0136

Reposted by Marco Ciapparelli

ERCbravenewword @ercbravenewword.bsky.social · 23/10/2025

A fascinating read in @theguardian.com on the psycholinguistics of swearing! Did you know Germans averaged 53 taboo words, while Brits e Spaniards listed only 16? Great to see the work of our colleague Simone Sulpizio & Jon Andoni Duñabeitia highlighted! 👏 www.theguardian.com/science/2025...

theguardian.com

Italian blasphemy and German ingenuity: how swear words differ around the world

Once swearwords were dismissed as a sign of low intelligence, now researchers argue the ‘power’ of taboo words has been overlooked

053

Reposted by Marco Ciapparelli

Melissa Franch, PhD @mfranch.bsky.social · 27/02/2025

I am incredibly proud to share my first, first-author paper as a postdoc with @benhayden.bsky.social . How does the human hippocampus, known for encoding concepts, represent the meanings of words while listening to narrative speech? www.biorxiv.org/content/10.1...

1231280

Reposted by Marco Ciapparelli

Marc Coutanche @marccoutanche.bsky.social · 26/08/2025

Here's a set of new results from my lab asking how the brain combines different ideas (concepts)! Now in press at J of Cog Neuro, we looked at how semantic composition (combining different concepts together) shapes brain activity. Preprint here: www.biorxiv.org/content/10.1... #neuroskyence

biorxiv.org

The Neural Consequences of Semantic Composition

Humans can create completely new concepts through semantic composition. These ‘conceptual combinations’ can be created by attributing the features of one concept to another (e.g., a lemon flamingo mig...

23513

Marco Ciapparelli @marcociapparelli.bsky.social · 18/07/2025

Compare concept representations across modalities in unimodal models, using the AlexNet convolutional neural network to represent images and an LLM to represent their captions

000

Marco Ciapparelli @marcociapparelli.bsky.social · 18/07/2025

Perform representational similarity analysis to compare how the same concepts are represented across languages (in their correponding monolingual models) and in different layers of LLMs

100

Marco Ciapparelli @marcociapparelli.bsky.social · 18/07/2025

Replace words with sense-appropriate and sense-inappropriate alternatives in the WiC annotated dataset and look at the effects of context-word interaction on embeddings and surprisal

100

Marco Ciapparelli @marcociapparelli.bsky.social · 18/07/2025

Extract word embeddings from BERT and inspect how context can modulate their representation. For example, what happens to "fruitless" when we place it in a sentence that points to its typical metaphorical meaning ("vain") as opposed to one where its meaning is literal ("without fruits")?

100

Marco Ciapparelli @marcociapparelli.bsky.social · 18/07/2025

I'm sharing a Colab notebook on using large language models for cognitive science! GitHub repo: github.com/MarcoCiappar... It's geared toward psychologists & linguists and covers extracting embeddings, predictability measures, comparing models across languages & modalities (vision). see examples 🧵

1114

Reposted by Marco Ciapparelli

Qihong (Q) Lu @qlu.bsky.social · 26/05/2025

I’d like to share some slides and code for a “Memory Model 101 workshop” I gave recently, which has some minimal examples to illustrate the Rumelhart network & catastrophic interference :) slides: shorturl.at/q2iKq code (with colab support!): github.com/qihongl/demo...

hidden state representation during training

1309

Marco Ciapparelli @marcociapparelli.bsky.social · 28/04/2025

7/7 Additional compositional representations emerge in left STS and semantic (but not compositional) representations in the left angular gyrus. Check out the preprint for more! Link to OSF project repo (includes code & masks used): osf.io/3dnqg/?view_...

osf.io

Characterizing semantic compositions in the brain: A model-driven fMRI re-analysis

Hosted on the Open Science Framework

100

Marco Ciapparelli @marcociapparelli.bsky.social · 28/04/2025

6/7 We find evidence of compositional representations in left IFG (BA45), even when focusing on a data subset where task didn't require semantic access. We take this to suggest BA45 represents combinatorial info automatically across task demands, and characterize combination as feature intersection

100

Marco Ciapparelli @marcociapparelli.bsky.social · 28/04/2025

5/7 We conduct confirmatory RSA in four ROIs for which we have a priori hypotheses of ROI-model correspondence (based on what we know of composition in models and what has been claimed of composition in ROIs), and searchlight RSAs in the general semantic network.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 28/04/2025

4/7 To better target composition beyond specific task demands, we re-analyze fMRI data aggregated from four published studies (N = 85), all employing two-word combinations but differing in task requirements.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 28/04/2025

3/7 To do so, we use word embeddings to represent single words, multiple algebraic operations to combine word pairs, and RSA to compare representations in models and target regions of interest. Model performance is then related to the specific compositional operation implemented.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 28/04/2025

2/7 Most neuroimaging studies rely on high-level contrasts (e.g., complex vs. simple words), useful to identify regions sensitive to composition, but less to know *how* constituents are combined (what functions best describe the composition they carry out)

100

Reposted by Marco Ciapparelli

Kanishka Misra @kanishka.bsky.social · 02/04/2025

another day another minicons update (potentially a significant one for psycholinguists?) "Word" scoring is now a thing! You just have to supply your own splitting function! pip install -U minicons for merriment

from minicons import scorer
from nltk.tokenize import TweetTokenizer

lm = scorer.IncrementalLMScorer("gpt2")

# your own tokenizer function that returns a list of words
# given some sentence input
word_tokenizer = TweetTokenizer().tokenize

# word scoring
lm.word_score_tokenized(
["I was a matron in France", "I was a mat in France"],
bos_token=True, # needed for GPT-2/Pythia and NOT needed for others
tokenize_function=word_tokenizer,
bow_correction=True, # Oh and Schuler correction
surprisal=True,
base_two=True
)

'''
First word = -log_2 P(word | <beginning of text>)

[[('I', 6.1522440910339355),
('was', 4.033324718475342),
('a', 4.879510402679443),
('matron', 17.611848831176758),
('in', 2.5804288387298584),
('France', 9.036953926086426)],
[('I', 6.1522440910339355),
('was', 4.033324718475342),
('a', 4.879510402679443),
('mat', 19.385351181030273),
('in', 6.76780366897583),
('France', 10.574726104736328)]]
'''

3217

Reposted by Marco Ciapparelli

Matilde Ellen Simonetti @matildellen.bsky.social · 25/03/2025

🚀 My first PhD paper is out! 🚀 "How do multiple meanings affect word learning and remapping?" was published in Memory & Cognition! Big thanks to my supervisors and co-authors (Iring Koch & @troembke.bsky.social). Curious? Read it here: rdcu.be/eeY9o #CognitivePsychology #WordLearning #Bilingualism

rdcu.be

How do multiple meanings affect word learning and remapping?

2144

Marco Ciapparelli @marcociapparelli.bsky.social · 19/03/2025

13/n In this context, LLMs flexibility allows to generate representations of possible/implicit meanings, which lead to representational drifts proportional to their plausibility. Data + code available: osf.io/s5edx/?view_...

osf.io

Conceptual Combination in Large Language Models: Uncovering Implicit Relational Interpretations in Compound Words with Contextualized Word Embeddings

Hosted on the Open Science Framework

000

Marco Ciapparelli @marcociapparelli.bsky.social · 19/03/2025

12/n Overall, our approach is consistent with theoretical proposals positing that word (and compound word) meaning should be conceptualized as a set of possibilities that might or might not be realized in a given instance of language use.

100

Marco Ciapparelli @marcociapparelli.bsky.social · 19/03/2025

11/n Also, bigger model != better: the best layer of BERT consistently outperformed the best layer of Llama. Results align with NLP/cognitive findings showing that LLMs are viable representational models of compound meaning but struggle with genuinely combinatorial stimuli.

110

Marco Ciapparelli @marcociapparelli.bsky.social · 19/03/2025

10/n Expectedly, LLMs vastly outperform DSMs on familiar compounds. Yet, unlike DSMs, LLM performance on novel compounds drops considerably. In fact, looking at novel compounds, some DSMs outperform the best layer of BERT and Llama! (image shows model fit; the lower the better).

110

Marco Ciapparelli @marcociapparelli.bsky.social · 19/03/2025

9/n As predicted, the closer a paraphrase CWE is to the original compound CWE, the more it was deemed plausible by participants. This holds for familiar compounds rated in isolation, familiar compounds rated in sentential contexts, and novel compounds rated in sentential contexts.

110

Marco Ciapparelli @marcociapparelli.bsky.social · 19/03/2025

8/n We reanalyzed possible relation task datasets using BERT-base (widely studied) and Llama-2-13b (representative of more recent, larger, and performant LLMs). As baselines, we used simpler (compositional) DSMs.

110