metadata
dataset_info:
features:
- name: original_id
dtype: int64
- name: is-augmented
dtype: string
- name: split
dtype: string
- name: question
dtype: string
- name: false_premise
dtype: bool
- name: num_hops
dtype: string
- name: fact_type
dtype: string
- name: Chinese
dtype: string
- name: Vietnamese
dtype: string
- name: Thai
dtype: string
- name: Indonesian
dtype: string
- name: Malay
dtype: string
- name: Khmer
dtype: string
- name: English
dtype: string
- name: Lao
dtype: string
splits:
- name: test
num_bytes: 1186415
num_examples: 1198
download_size: 556876
dataset_size: 1186415
configs:
- config_name: default
data_files:
- split: test
path: data/test-*
language:
- multilingual
task_categories:
- question-answering
This dataset is from the paper Analyzing LLMs' Knowledge Boundary Cognition Across Languages Through the Lens of Internal Representations. It is designed to evaluate how well LLMs recognize the boundaries of their knowledge across multiple languages.
The dataset contains questions in English and several other languages (Chinese, Vietnamese, Thai, Indonesian, Malay, Khmer, and Lao), along with information about whether the question is designed to test knowledge boundaries (e.g., whether it contains a false premise).