Request for Little Prince (LPP) Narratives File

[Jeryi-Sun]

Hello,

First of all, thank you for creating and sharing this library — it’s incredibly helpful!

I noticed that the only LeBel file available for download is assembly_lebel_uts03.pkl.
Could you please share the Narratives LeBel file for Little Prince (LPP) as well?

Thank you very much! 🙏

[TahaBinhuraib]

Hi! Thank you so much!

We're currently working on a scalable way to host other datasets with more datasets and more subjects, and will let you know as soon as possible. In the meantime, could you follow the steps outlined in the https://github.com/GT-LIT-Lab/litcoder_release repo to preprocess your own data?

Please let me know if you have any other questions!

We will also be releasing youtube video with tutorials on how to setup litcoder_release for easier use!

[TahaBinhuraib]

Once you have all the data, you can then proceed to generate your own assemblies instead of loading pre assembled packages!

[Jeryi-Sun]

Hi @TahaBinhuraib, thank you very much for the quick reply!

I’m relatively new to this area, and I’m still getting familiar with the neuroscience-related datasets. I’m a bit worried that I might make mistakes during the preprocessing steps, which could lead to incorrect assemblies or difficulty reproducing the paper’s results.

If it’s possible to upload the processed Narratives file or Little Prince (LPP), it would greatly help me validate my pipeline and ensure I’m following the correct procedures. I would be extremely grateful!

Thank you again for your work and for considering this request!

[TahaBinhuraib]

Of course! Our goal is to try and maintain this repo to the best of our abilities, and are willing to help all the way!

I'll try to have a couple of more assemblies uploaded sometime this weekend, or the next! I'll keep you posted on progress!

[Jeryi-Sun]

Thanks so much for the quick response — really appreciate it!

That would be incredibly helpful. Having the additional assemblies (Narratives or LPP) would make it much easier for me to verify my preprocessing pipeline and ensure I’m reproducing the results correctly. Thank you for taking the time to support users like me who are still getting familiar with this area.

Looking forward to your update — and thanks again for maintaining the repo so thoroughly! 🙏

[Jeryi-Sun]

Hi @TahaBinhuraib, sorry to bother you again — and thank you very much for all your help so far!

I wanted to check one more thing:
Currently, the repo provides the LeBel assembly for only one subject (assembly_lebel_uts03.pkl).
Would it be possible in the future to also release LeBel assemblies from other users/subjects, if licensing and storage allow?

Having a few additional subjects (even just 1–2 more) would be extremely helpful for validating cross-subject consistency and ensuring that my preprocessing and assembly pipeline are correct.

Totally understand if this is limited by data-sharing constraints — just wanted to ask in case it’s possible.
Thank you again for your time and all the effort you’re putting into maintaining this project! 🙏

[TahaBinhuraib]

Hi!

Apologies for the delay! Storage has been an issue as of late... My current estimate is to have 1-2 more assemblies sometime next Thursday! Once again, sorry for the delay!

[Jeryi-Sun]

Hi @TahaBinhuraib, thank you again for the update — and no worries at all about the delay!

I wanted to ask one more question regarding the storage limitations you mentioned:

If storage is the main bottleneck, would it be possible to provide text-only versions of the assemblies (i.e., without the audio files)?
This would apply not only to the Narratives/LPP assemblies, but also — and even more importantly — to additional LeBel users.

For many language-model experiments (RSA, voxel-wise encoding, parcel-level analyses, etc.), the audio is not required; the text alone is fully sufficient.
A text-only release would be much smaller in size and could make it feasible to share 1–2 more LeBel users without hitting storage limits.

Even partial or temporary text-only assemblies would be extremely helpful for validating cross-subject consistency and ensuring that our preprocessing pipelines are correct before we move on to full audio-aligned experiments.

Thank you again for considering this, and for all the effort you’re putting into maintaining this project — it’s incredibly appreciated! 🙏

[Jeryi-Sun]

Hi @TahaBinhuraib, hope you’re doing well — sorry to bother you again, and thank you very much for all the updates so far!

I just wanted to ask a quick clarification related to the storage constraints you mentioned.
If storage is the main bottleneck, would it be possible to release only the fsaverage5-space parcel / label-level activations (e.g., parcel-wise or ROI-wise time series), without the raw audio, voxel-wise data, or text?

For many downstream analyses (e.g., parcel-level RSA, cross-subject consistency checks, atlas-level alignment), having activations already projected to fsaverage5 labels would be fully sufficient, and the storage footprint should be much smaller than full voxel-wise assemblies.

Even a very minimal release (e.g., 1–2 additional LeBel subjects, labels only) would already be extremely helpful for validating cross-subject robustness and ensuring our preprocessing pipeline matches yours.

Totally understand if this is not feasible at the moment — just wanted to ask in case this option makes things easier.
Thanks again for all the effort you’re putting into maintaining the repo, it’s really appreciated!

Best regards,
Jeryi

[TahaBinhuraib]

Hi Jeryi,

Apologies for the delayed reply, and thank you so much for your kind words about the repo; we really appreciate it!

Generating parcel- or ROI-level activations would take some extra work. In our analyses, we usually fit voxel-wise models and then average correlations within each region to get a performance metric, rather than averaging activations directly. Averaging activations can also be misleading because it can wash out fine-grained information: different voxels in the same region may respond differently, and taking the mean can reduce signal specificity and underestimate the true effect. Usually people who work with vision datasets, such as NSD, do this, but I haven't seen anyone do it for language yet.

We’re still figuring out a scalable way to share data without extra preprocessing. In the meantime, I’ve made subject UTS02 from the LeBel dataset available here: https://drive.google.com/file/d/1Y8DXu6XSUtnPE6DI_Yqoe1FpQknrYVuM/view?usp=sharing

This should work exactly the same way subject UTS03 works. Please let me know if you encounter any problems!

I'll add one subject from the narratives dataset over the weekend!

Hope this helps!

[Jeryi-Sun]

Hi Taha,

Thank you so much for the detailed explanation — that’s very helpful, and I really appreciate the clarification on the voxel-wise modeling choices.

Thanks as well for sharing UTS02; this is extremely helpful. I’ll start working with it and will let you know if I run into any issues.

Really appreciate you also adding a Narratives subject — looking forward to it.
Thanks again for all your support and for maintaining the repo!

Best regards,
Jeryi

[TahaBinhuraib]

Hi Jeryi,

Hope all is well!

I was able to upload subject 249 from the Narratives data as an assembly. Please let me know if you have any problems using it!

Assembly link: https://drive.google.com/file/d/1YPV3Mhj1h-QimKs5f57kxNTu0qBb9scq/view?usp=drive_link

Best regards,
Taha