Language Learning Aptitude (ds003508)

Brain MRI from a Lund University study of how individual differences in the ability to learn a foreign language relate to brain structure. Young adults completed the LLAMA aptitude tests for vocabulary learning, phonetic memory and grammatical inference, and were scanned at 7 T. The study looked at cortical thickness and surface area from the sub-millimetre anatomical scans and at white matter microstructure from diffusion kurtosis imaging in language tracts. The data are on OpenNeuro in BIDS format under CC0.

Composition

57 right-handed university students, 42 women and 15 men, aged 20 to 27 with a mean of 22.7 years. According to the paper, 35 grew up in monolingual German-speaking families and 22 in monolingual Swedish-speaking families. The participants file adds the three LLAMA subtest scores, an operation span score, a Raven matrices score, the Goldsmiths musical sophistication index and the self-reported number of languages learned. There is no patient group.

Acquisition

All scans come from a 7 T Philips Achieva at the Lund University Bioimaging Center with a 32-channel receive head coil and dielectric pads. Each participant has a 3D T1-weighted gradient echo scan at 0.8 mm isotropic and a matching low flip angle proton density weighted scan at 1.6 mm isotropic, the latter used to correct the T1-weighted image for intensity bias. The diffusion protocol has a single-shell run with 56 directions at b = 2000 s/mm² and 2 mm voxels for tractography, a reversed phase-encoding b = 0 run for distortion correction, and a multi-shell run with b-values of 100, 700, 1400 and 2000 s/mm² at 2 × 2 × 4 mm for kurtosis estimation.

Annotations

There are no image labels. Behavioural test scores per participant are the only non-imaging data.

Known limitations

The sample is small, young and mostly female, and all participants are students. Two participants (sub-003 and sub-045) have no diffusion files; the paper attributes missing tractography data in two participants to scanner problems. The paper excluded further participants from individual analyses for poor cortical reconstruction or diffusion signal loss, so its analysed sample sizes are below 57. The dataset README still says the paper will be added once published.