BrainAlign/brain-lm-alignment-ds002236

Brain–language-model alignment: ds002236

Lytle et al. 2020 — orthographic, phonological and semantic word processing in school-aged children (8.7–15.5), auditory and visual.

• Paper: https://pubmed.ncbi.nlm.nih.gov/31956678/
• Data: https://openneuro.org/datasets/ds002236/versions/1.0.1
• Generated: 2026-08-28
• Pipeline: https://github.com/suchirsalhan/cdl-representations-brains-babylms

Read this first: does the measurement work?

Every alignment number in this dataset is only as meaningful as the brain RDMs it was computed against. So before any model result, the same pipeline is asked whether anything stimulus-driven correlates with those RDMs — stimulus duration, intensity, word length, frequency, phoneme and syllable counts, an acoustic model of the audio where the stimuli are audio, and the study's own condition contrast — each tested by a permutation test that shuffles stimulus identity.

GATE: FAILED. 0/6 stimulus tests are significant after Holm correction — not the acoustic model of the audio the children actually heard, not the study's own experimental contrast.

The alignment numbers below are therefore uninterpretable as evidence about language models. They measure a representational geometry that does not demonstrably encode the stimuli. They are published for completeness and for whoever fixes the estimator, not as a result. Do not cite them as evidence that models fail to align with the developing brain.

Measured cause, from control/:

• RDM effective rank: 54 of 72 stimuli

Note that this is NOT ds003604's failure mode. There, the RDM effective rank was ~3 of 40-48 stimuli -- near-degenerate betas that could not express stimulus-level structure at all. The rank recorded above is a large fraction of the stimulus count, so these RDMs do carry stimulus structure and the control failing here means the specific controls tested did not reach significance, not that the measurement is uninterpretable. Check control/ for which controls ran: an acoustic or visual control needs the dataset's stimulus files present, and reports zero features if they are not.

What was built

6 task × session cells, each an RDM over the stimuli shared by that cell's subjects, with voxel patterns z-scored within run before aggregation (without that, the RDM measures scanner drift rather than language) and an inter-subject noise ceiling.

task	session	n_stim	ceiling_lower	ceiling_upper	ceiling_n
Phon	ses-11+	96	0.302747	0.40391	20
Phon	ses-11	96	0.234711	0.411819	10
Phon	ses-9	96	0.255574	0.441516	9
Sem	ses-11+	48	0.436333	0.506525	21
Sem	ses-11	48	0.390186	0.542448	8
Sem	ses-9	48	0.344433	0.482067	11

Model grid: 15 families, 524 alignment rows across 2 cells.

mean noise ceiling	0.327
best alignment anywhere	0.1139
as a fraction of ceiling	44.5%
families equivalent to zero (TOST ±0.05)	0/15
Pythia scale trend	ρ = +0.148, p = 0.68

Per family

family	n_checkpoints	rsa_mean	rsa_sd	rsa_abs_max	frac_of_ceiling_abs_max	p_equivalence_tost
babylm-gpt2	9	0.0287	0.0059	0.0534	0.1689	nan
pico-decoder-medium	21	0.0276	0.0346	0.1139	0.4455	nan
pythia-1b-full	21	0.0217	0.0268	0.0762	0.298	nan
pico-decoder-small	21	0.0213	0.0167	0.1044	0.4086	nan
babylm-gpt2-7	9	0.0207	0.0065	0.0445	0.1292	nan
pythia-410m-full	21	0.0188	0.0139	0.0741	0.2901	nan
babylm-gpt2-5	9	0.0186	0.0034	0.0368	0.116	nan
babylm-gpt2-3	9	0.018	0.007	0.0432	0.1346	nan
pico-decoder-large	21	0.0179	0.0334	0.0796	0.3116	nan
beetle-fineweb3-eng	19	0.0177	0.0213	0.0778	0.3044	nan
pythia-160m-full	21	0.0155	0.012	0.0716	0.2801	nan
pico-decoder-tiny	21	0.014	0.0121	0.0758	0.2967	nan
pythia-1.4b-full	21	0.0125	0.0271	0.0946	0.37	nan
pythia-70m-full	21	0.0121	0.0155	0.0753	0.2947	nan
beetle-humanscale-eng	18	0.0101	0.006	0.0477	0.1865	nan

Dataset-specific notes

The accession is not stated in the data article; it was resolved to ds002236 by matching OpenNeuro's own dataset name ("Cross-Sectional Multidomain Lexical Processing") AND the per-subject age range in participants.tsv (8.67–15.5) against the range the article reports. Best developmental axis of the four datasets: explicit per-subject age at scan, continuous rather than binned. Six tasks crossing modality (auditory/visual) with judgement (rhyme/spelling/semantic) — a modality control no other dataset here provides. A third of trials are coded null (Tones/nullsilence.WAV) and are excluded from the stimulus set.

Files

path	what
alignment_by_checkpoint.csv	every model × checkpoint × cell, with ceiling
alignment_by_family.csv	per family, with equivalence tests
alignment_by_cell.csv	per task × session
ceilings_*.csv	noise ceiling per cell
control/	the positive control and RDM dimensionality — the gate
scale_ladder.csv	the Pythia 70M→1.4B scale test
fig_*.pdf, fig_*.png	figures

Method

Representational similarity analysis. For each cell, a brain RDM over stimuli (correlation distance between per-stimulus GLM beta patterns, within-run z-scored, aggregated across subjects) is compared by Spearman correlation with a model RDM over the same stimuli, taken from each checkpoint's hidden states. Alignment is reported raw and as a fraction of the inter-subject noise ceiling, and judged against a null built from the PARC suite — 18 models differing only by random seed, which is what 'no effect' looks like on this measurement.

Null and fixation trials are excluded from the stimulus set. For paired designs the stimulus identity is the pair, not either word alone.