| ## 📊 Guwen Datasets |
|
|
| ### 1. Erya Benchmark |
| We utilize the translation subset of the **Erya** benchmark as our foundational data. For more details, please refer to the [Erya Official Repository](https://github.com/RUCAIBox/Erya). |
|
|
| ### 2. Robust-Erya Test Set (Proposed) |
| To systematically evaluate model robustness, we introduce **Robust-Erya**, an extension of the Erya benchmark incorporating diverse linguistic noises across multiple domains. |
|
|
| #### Data Scale & Diversity |
| Robust-Erya covers **5 distinct domains** from the Erya test suite: |
| - `han`: Book of Han |
| - `xint`: New Tang History |
| - `ming`: Ming History |
| - `taip`: Taiping Guangji |
| - `xux`: Xu Xiake's Travels |
|
|
| #### Noise Taxonomy |
| For each domain, we provide **3 types of noise** with **5 intensity levels** (ranging from 0.1 to 0.5 ratio): |
|
|
| | Noise Type | Description | File Path Example | |
| | :--- | :--- | :--- | |
| | **sub** | (a) Substitution Noise | `dataset/Robust-Erya/han/sub/han_sub_0.1.json` | |
| | **ins** | (b) Redundancy Noise | `dataset/Robust-Erya/han/ins/han_ins_0.3.json` | |
| | **del** | (c) Omission Noise | `dataset/Robust-Erya/han/del/han_del_0.5.json` | |
|
|
| Total evaluation files: 5 (domains) × 3 (noise types) × 5 (ratios) = **75 test files**. |
