|
|
--- |
|
|
language: |
|
|
- zh |
|
|
- en |
|
|
pretty_name: HFTP Experimental Datasets |
|
|
tags: |
|
|
- syntactic-analysis |
|
|
- neuroscience |
|
|
- brain-model-alignment |
|
|
- cross-linguistic |
|
|
- NeurIPS2025 |
|
|
task_categories: |
|
|
- text-classification |
|
|
- token-classification |
|
|
size_categories: |
|
|
- n<1K |
|
|
--- |
|
|
|
|
|
# HFTP Experimental Datasets |
|
|
|
|
|
## Overview |
|
|
|
|
|
This dataset contains experimental corpora used in the paper **"Hierarchical Frequency Tagging Probe (HFTP): A Unified Approach to Investigate Syntactic Structure Representations in Large Language Models and the Human Brain"**, published at **NeurIPS 2025**. |
|
|
|
|
|
📄 **Paper**: [https://arxiv.org/abs/2510.13255](https://arxiv.org/abs/2510.13255) |
|
|
|
|
|
💻 **Code Repository**: [https://github.com/LilTiger/HFTP](https://github.com/LilTiger/HFTP) |
|
|
|
|
|
## Dataset Description |
|
|
|
|
|
These datasets are designed for analyzing syntactic structure processing in large language models (LLMs) using the Hierarchical Frequency Tagging Probe (HFTP) method. The corpora include both syntactic and natural language data in Chinese and English. All files are in CSV format with one sentence per line. |
|
|
|
|
|
|
|
|
## Files |
|
|
|
|
|
### Syntactic Corpora |
|
|
|
|
|
- **`Chinese_syntactic_corpus.csv`**: Chinese four-syllable syntactic corpus for extracting syntactic neural representations in LLMs |
|
|
- **Example**: "老牛耕地", "朋友请客" |
|
|
|
|
|
- **`English_syntactic_corpus.csv`**: English four-word syntactic corpus with parallel design for cross-linguistic syntactic analysis |
|
|
- **Example**: "fat rat sensed fear", "wood shelf holds cans" |
|
|
|
|
|
### Natural Language Corpora |
|
|
|
|
|
#### Chinese Natural Language Data |
|
|
|
|
|
- **`Chinese_8-natural.csv`**: 8-character Chinese natural language corpus containing diverse text types (everyday dialogue, news reports, literary excerpts, poetry) |
|
|
- **Example**: "森林火势得到控制。", "列车准点抵达站台。" |
|
|
|
|
|
- **`Chinese_9-natural.csv`**: 9-character Chinese natural language corpus with extended samples for frequency analysis |
|
|
- **Example**: "临床试验数据公布了。", "姐姐，这裙子有蓝色吗？" |
|
|
|
|
|
- **`Chinese_8-zhwiki.csv`**: 8-character Chinese Wikipedia corpus providing encyclopedia-style natural language data |
|
|
- **Example**: "这个发现归功于他。", "人身牛首，長於姜水。" |
|
|
|
|
|
#### English Natural Language Data |
|
|
|
|
|
- **`English_8-naturale.csv`**: 8-word English natural language corpus containing everyday dialogue, news reports, and literary prose |
|
|
- **Example**: "With malice toward none, with charity toward all.", ""Tonight, shall we watch the meteor shower together?"" |
|
|
|
|
|
- **`English_9-naturale.csv`**: 9-word English natural language corpus with extended samples |
|
|
- **Example**: "Clinical trial data were publicly released this morning nationwide.", "Shall we share sushi together tonight by the river?" |
|
|
|
|
|
- **`English_8-enwiki.csv`**: 8-word English Wikipedia corpus extracted from English Wikipedia articles |
|
|
- **Example**: "World and Its Peoples: Eastern and Southern Asia.", "Cognitive systems engineering: New wine in new bottles." |
|
|
|
|
|
|
|
|
## Citation |
|
|
|
|
|
If you use these datasets in your research, please cite: |
|
|
```bibtex |
|
|
@misc{an2025hierarchicalfrequencytaggingprobe, |
|
|
title={Hierarchical Frequency Tagging Probe (HFTP): A Unified Approach to Investigate Syntactic Structure Representations in Large Language Models and the Human Brain}, |
|
|
author={Jingmin An and Yilong Song and Ruolin Yang and Nai Ding and Lingxi Lu and Yuxuan Wang and Wei Wang and Chu Zhuang and Qian Wang and Fang Fang}, |
|
|
year={2025}, |
|
|
eprint={2510.13255}, |
|
|
archivePrefix={arXiv}, |
|
|
primaryClass={cs.CL}, |
|
|
url={https://arxiv.org/abs/2510.13255}, |
|
|
} |
|
|
``` |
