lapa-llm/impossible-questions

---
dataset_info:
features:
- name: id
dtype: int64
- name: question
dtype: string
- name: category
dtype: string
- name: explanation
dtype: string
- name: correct_answer
dtype: string
- name: conversations
list:
- name: from
dtype: string
- name: value
dtype: string
splits:
- name: train
num_bytes: 55935
num_examples: 135
download_size: 26382
dataset_size: 55935
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: mit
---

# Dataset Card for Impossible Questions

## Dataset Description

**Dataset Summary**

This dataset is a compilation of the questions, that can't be answered from the question text. It is intended to have negative answer from the model. Besides, each entry contains explanation of the answer. Here is an example of one entry:

```
>> Повзе 4 равлики, біля них 5 дітей. Скільки у них загалом дядьків?
Неможливо визначити
```

The goal of this dataset is to teach the model to recognize when there is no correct answer in the text of the question, and to teach the model to answer correctly in such cases.

<!--[Provide a brief overview of your dataset - what it contains, its purpose, and why it was created. Example: "This dataset contains X examples of Ukrainian text collected from Y sources, designed to support the development of Ukrainian language models."] -->

**Languages**

- Ukrainian (uk)

<!-- **Dataset Structure** -->

<!-- The dataset is organized into the following splits:

| Split | Examples |
|-------|----------|
| Train | [number] |
| Validation | [number] |
| Test | [number] | -->

**Data Fields**

- `question`: The unanswerable question text posed to the model
- `category`: Classification or type of unanswerable question
- `explanation`: Explanation of why the question cannot be answered from the given information
- `correct_answer`: The appropriate response indicating the question is unanswerable
- `conversations`: Dialog version of the dataset for LLM training

<!-- ## Dataset Creation

**Source Data**

Preprocessed data comes from https://huggingface.co/datasets/omarkamali/wikipedia-monthly, latest.uk split. -->

<!-- **Data Collection Process**

[Explain how the data was collected and any processing steps applied]

<!-- **Annotations**

[If applicable, describe any annotation process, who annotated, annotation guidelines, etc.] -->

## Considerations for Using the Data

**Social Impact**

This dataset was created to support Ukrainian language AI development and improve language technology accessibility for Ukrainian speakers.

<!-- **Bias and Limitations**

[Discuss any known biases, limitations, or potential issues with the dataset. Be transparent about what the dataset may not be suitable for.] -->

**Recommendations**

You can use this dataset for the following purposes:

- Question answering

## Citation

TBD

<!--
**BibTeX**

```bibtex
@dataset{dataset_name,
author = {[Your Name/Organization]},
title = {[Dataset Name]},
year = {2025},
publisher = {HuggingFace},
url = {https://huggingface.co/datasets/[your-org]/[dataset-name]}
}
```
-->

## Contact

<!-- For questions or feedback, please contact [your contact information] or open an issue on the dataset repository. -->

For questions or feedback, please open an issue on the dataset repository.

## License

MIT License

---

*This dataset is part of the "Lapa" - Ukrainian LLM initiative to advance natural language processing for the Ukrainian language.*

数据集信息：
特征字段：
- 名称：id，数据类型：int64（64位整数）
- 名称：question，数据类型：string（字符串）
- 名称：category，数据类型：string（字符串）
- 名称：explanation，数据类型：string（字符串）
- 名称：correct_answer，数据类型：string（字符串）
- 名称：conversations（对话列表），数据类型为列表：
- 子字段：from（对话来源），数据类型：string（字符串）
- 子字段：value（对话内容），数据类型：string（字符串）
数据划分：
- 划分名称：train（训练集），字节数：55935，样本数：135
下载大小：26382，数据集大小：55935
配置项：
- 配置名称：default（默认配置），数据文件：
- 数据划分：train（训练集），文件路径：data/train-*
许可证：MIT

## 「无法回答问题」数据集卡片

### 数据集说明

#### 数据集概览
本数据集汇集了无法通过题干文本推导答案的问题，旨在训练模型生成否定性回答。此外，每条数据均包含对应答案的解释说明。以下为单条数据示例：

>> 有4只兔子，旁边有5个小孩。他们总共有多少位舅舅？
无法确定

本数据集的目标是让模型识别题干中不存在有效可解信息的场景，并在这类场景下给出正确的应对回应。

#### 语言
- 乌克兰语（uk）

#### 数据字段
- `question`：提交给模型的无法作答的问题文本
- `category`：无法作答问题的分类或类型
- `explanation`：说明为何无法通过给定信息回答该问题的解释文本
- `correct_answer`：表明该问题无法作答的标准回应
- `conversations`：适配大语言模型（Large Language Model，LLM）训练的对话格式数据集版本

### 数据使用注意事项

#### 社会影响
本数据集旨在推动乌克兰语人工智能（AI）发展，提升乌克兰语使用者的语言技术可及性。

#### 使用建议
本数据集可用于以下场景：
- 问答任务

### 引用信息
待公布（TBD）

### 联系方式
如有疑问或反馈，请在数据集仓库中提交issue。

### 许可证
MIT许可证

本数据集隶属于「Lapa」项目——一项旨在推进乌克兰语自然语言处理发展的乌克兰语大语言模型倡议。