Genius. "key" is a context dependent abstract concept after all. who is to say what is the real key?

LOL. Creative!

Many congratulations!

Many congratulations to @sophiearheix.bsky.social for her K99/R00 award! Her work aims to apply TMS to novel targets for aphasia. Look out for exciting papers from her new lab!

The Two Word Test as a semantic benchmark for large language models - Scientific Reports Scientific Reports - The Two Word Test as a semantic benchmark for large language models

Nice work! You may want to take a look this paper, which is tangentially related. We find that LLMs are "too creative" in making sense of word combinations (relative to humans). www.nature.com/articles/s41...

Announcing 2026 Conference at the Institute of Mind and Brain, U Of South Carolina

A model of Frame and Verb Compliance in language acquisition, 2007 Neurocomputing

btw, this is a tiny model that shows item-based learning (memorizing sentence frames) initially and then switches to lexical cues. from the good old days when networks had 3-4 layers.
www.sciencedirect.com/science/arti...

Very nice work. Reminds me of "item-based learning" proposal by tomasello and others with respect to language acquisition in children. how kids memorize word combinations that lead to full generalization later.

We have, however, used them to model errors in picture naming. They produce various types of errors seen in aphasic patients (eg, semantic, phonological, nonword, neologism, mixed errors). I would be skeptical that we can extend this to past tense but we haven't tried yet.

The issue is that LLMs have a relatively poor representation of phonology. They have tokens that are sort of like morphemes, but they don't have a full fledged phonology. This is kind of important for past tense. That's why it is hard to use them for modeling sublexical processes.

"I asked Siri". There's your first mistake.

Crazy good if you drew it blind.

Finally some snow in SC to the delight of kids. Quicky taking advantage in the mornimg before it melts in the afternoon.

Legos are usually a safe bet.

We found similar differentiation (eg fig 1b) between people and places using written names (as opposed to pictures). these areas also overlap with general semantic network.

Proper and common names in the semantic system - PubMed Proper names are an important part of language and communication. They are thought to have a special status due to their neuropsychological and psycholinguistic profile. To what extent proper names rely on the same semantic system as common names is not clear. In an fMRI study, we presented the same …

Very nice work, congratulations. You may find this paper relevant:

pubmed.ncbi.nlm.nih.gov/36372812

😀

Is Representational Similarity Analysis Reliable? A Comparison with Regression Representational Similarity Analysis (RSA) is a popular method for analyzing neuroimaging and behavioral data. Here we evaluate the accuracy and reliability of RSA in the context of model selection, a...

Representation Similarity Analysis (RSA) is a popular method for analysis of neuroimaging and behavioral data. Our recent paper suggests it has 'issues' and has potential to produce misleading results.
arxiv.org/abs/2511.00395

Peter and Bonnie McCausland Distinguished Chair in Neuroscience

🔥🔥New faculty position in lifespan neuroimaging (McCausland Distinguished Chair). That's an amazing opportunity to join our brand-new Brain Health Network in Columbia. If you are interested in, please DM me.
uscjobs.sc.edu/postings/190...

Need more subjects.

Good luck for the surgery! Exemplary move to turn this into a rich study.

In addition to the 7 jobs in brain/language at University of South Carolina, we also have a philosophy job on issues related to AI
philjobs.org/job/show/29754

Unsupported client – Canva Unsupported client – Canva

Direct links to each position

www.canva.com/design/DAGxX...

Seven faculty positions in language/brain health at University of South Carolina!
sc.edu/study/colleg...

He ran an excellent journal club in the evenings during my student days at IU. Fondly remember "how far you can go without representations" discussions and more. He will be missed.

If your work involves use of properties of English words, you are likely missing something if you are not using SCOPE.

New variables added in this version include gpt4o ratings, human motivation, iconicity, cognition, and spelling-sound transparency ratings.

New version of SCOPE is out! It is the largest metabase of psycholinguistic properties of words.
sc.edu/study/colleg...

IU has been a leader in cognitive science for a long time. Sad to see this elimination, and so many others, as an alum.