EQ-bench_ca

# Dataset Card for EQ Bench Dataset (Catalan Version)

This dataset card documents the Catalan adaptation of the EQ-Bench benchmark. The original dataset was designed to evaluate emotional reasoning in language models through dialogue-based prompts.

## Dataset Details

### Dataset Description

EQ-Bench (Catalan Version) is a translated and linguistically adapted version of the original EQ-Bench dataset.
Its design responds to the need to adapt the emotional detection capabilities of multilingual models, recognizing that the expression and perception of emotions varies significantly across languages.

Key adaptations include:
1) The conversion of adjectival emotion labels into nominal forms to resolve gender agreement ambiguity.
2) The unification of semantically equivalent labels that appeared in different grammatical forms in the original dataset (e.g., *pride/proud* → *orgull*).
3) The replacement of Anglo-Saxon proper names with culturally appropriate Catalan ones to maintain linguistic coherence. Additionally, definite articles (e.g., el, la, l’) have been added following Catalan grammatical norms, depending on the gender and initial sound of the name.

- **Curated by:** Barcelona Supercomputing Center (BSC)
- **Funded by:** [AINA](https://projecteaina.cat/) ; [ILENIA](https://proyectoilenia.es/)
- **Language(s) (NLP):** ca (Catalan)
- **License:** CC BY 4.0

### Dataset Sources

- **Repository:** [EQ-Bench](https://huggingface.co/datasets/pbevan11/EQ-Bench)
- **Paper:** Paech, S. J. (2023). *EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models*. [arXiv:2312.06281](https://arxiv.org/abs/2312.06281)

## Uses

### Direct Use

This dataset can be used to:
- Evaluate emotional reasoning in Catalan-language LLMs
- Study multilingual performance variation in emotion understanding
- Fine-tune or test classification models for emotion recognition from dialogue

### Out-of-Scope Use

This dataset is not intended for:
- Training general-purpose sentiment analysis systems without considering the emotion-specific context of dialogues.
- Applications such as real-time mental health diagnostics or therapeutic interventions.
- Generating outputs in legal or clinical contexts without human oversight.
- Use in non-Catalan contexts, as the dataset has been culturally and linguistically localized.

## Dataset Structure

### Dataset Structure

Each entry in the dataset follows this structure:
```json
{
"{
"prompt": "...",
"reference_answer": { // The full prompt for the model, including the dialogue, emotion options, and formatting instructions
"emotion1": "...", // First emotion label
"emotion2": "...", // Second emotion label
"emotion3": "...", // Third emotion label
"emotion4": "...", // Fourth emotion label
"emotion1_score": int, // Annotated scores
"emotion2_score": int,
"emotion3_score": int,
"emotion4_score": int
},
"reference_answer_fullscale": {
"emotion1": "...", //Same emotion labels
"emotion2": "...",
"emotion3": "...",
"emotion4": "...",
"emotion1_score": int, // Resolution scores
"emotion2_score": int,
"emotion3_score": int,
"emotion4_score": int
}
}
}
```

## Dataset Creation

### Curation Rationale

The design of this dataset responds to the need to adapt the emotional detection capabilities of multilingual models, recognizing that the expression and perception of emotions varies significantly across languages.

### Source Data

EQ Bench original dataset: (https://huggingface.co/datasets/pbevan11/EQ-Bench)

#### Who are the source data producers?

All credits go to the creator of the original EQ Bench dataset, Samuel J. Paech.

## Citation

**APA:**

Paech, S. J. (2023). EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models. arXiv. https://arxiv.org/abs/2312.06281

## More information

This work/research has been promoted and financed by the Government of Catalonia through the [Aina project](https://projecteaina.cat/).
This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the [project ILENIA](https://proyectoilenia.es/) with reference 2022/TL22/00215337

## Contact point

Language Technologies Unit ([e-mail]) at the Barcelona Supercomputing Center (BSC).

# EQ基准数据集（加泰罗尼亚语版）数据集卡片

本数据集卡片旨在介绍EQ-Bench基准测试的加泰罗尼亚语适配版本。原始数据集旨在通过基于对话的提示词，评估语言模型的情绪推理能力。

## 数据集详情

### 数据集描述
EQ-Bench（加泰罗尼亚语版）是原始EQ-Bench数据集经过翻译与语言适配后的版本。其开发旨在适配多语言模型的情绪检测能力，因为不同语言间情绪的表达与感知存在显著差异。

关键适配项包括：
1) 将形容词性情绪标签转换为名词形式，以解决性别一致歧义。
2) 统一原始数据集中以不同语法形式出现的语义等价标签（例如*pride/proud* → *orgull*）。
3) 用符合文化适配的加泰罗尼亚语专有名词替换盎格鲁-撒克逊式专有名词，以保持语言连贯性。此外，根据加泰罗尼亚语语法规范，依据名词的性别与首音添加定冠词（如el、la、l’）。

- **数据整理方:** 巴塞罗那超级计算中心（Barcelona Supercomputing Center, BSC）
- **资助方:** [AINA](https://projecteaina.cat/) ; [ILENIA](https://proyectoilenia.es/)
- **NLP支持语言:** 加泰罗尼亚语（ca）
- **许可协议:** CC BY 4.0

### 数据集来源
- **代码仓库:** [EQ-Bench](https://huggingface.co/datasets/pbevan11/EQ-Bench)
- **论文:** Paech, S. J. (2023). *EQ-Bench：面向大语言模型（Large Language Model, LLM）的情绪智能基准测试*. [arXiv:2312.06281](https://arxiv.org/abs/2312.06281)

## 使用场景

### 直接使用场景
本数据集可用于：
- 评估加泰罗尼亚语大语言模型的情绪推理能力
- 研究多语言模型在情绪理解任务中的性能差异
- 针对对话情绪识别任务，微调或测试分类模型

### 超出适用范围的场景
本数据集不应用于：
- 未考虑对话特定情绪上下文的通用情感分析系统训练
- 实时心理健康诊断或治疗干预类应用
- 无人工监督的法律或临床场景内容生成
- 非加泰罗尼亚语语境下的使用，因为本数据集已完成文化与语言本地化适配。

## 数据集结构

### 数据集结构
数据集中的每个条目遵循如下格式：
json
{
"{
"prompt": "...",
"reference_answer": { // 模型完整提示词，包含对话、情绪选项与格式说明
"emotion1": "...", // 首个情绪标签
"emotion2": "...", // 第二个情绪标签
"emotion3": "...", // 第三个情绪标签
"emotion4": "...", // 第四个情绪标签
"emotion1_score": int, // 人工标注得分
"emotion2_score": int,
"emotion3_score": int,
"emotion4_score": int
},
"reference_answer_fullscale": {
"emotion1": "...", // 同上述情绪标签
"emotion2": "...",
"emotion3": "...",
"emotion4": "...",
"emotion1_score": int, // 全量参考得分
"emotion2_score": int,
"emotion3_score": int,
"emotion4_score": int
}
}
}

## 数据集构建

### 构建初衷
本数据集的开发旨在适配多语言模型的情绪检测能力，因为不同语言间情绪的表达与感知存在显著差异。

### 源数据
EQ Bench原始数据集：(https://huggingface.co/datasets/pbevan11/EQ-Bench)

#### 源数据提供者是谁？
所有荣誉归于原始EQ Bench数据集的创作者Samuel J. Paech。

## 引用格式
**APA:**
Paech, S. J. (2023). EQ-Bench：面向大语言模型的情绪智能基准测试. arXiv. https://arxiv.org/abs/2312.06281

## 更多说明
本研究由加泰罗尼亚政府通过[Aina项目](https://projecteaina.cat/)推广并资助。本工作由西班牙数字化与公共职能部资助，由欧盟下一代欧盟（NextGenerationEU）框架下的[ILENIA项目](https://proyectoilenia.es/)资助，项目编号2022/TL22/00215337。

## 联系方式
巴塞罗那超级计算中心（BSC）语言技术部门（[e-mail]）。