---
dataset_info:
  features:
    - name: id
      dtype: string
    - name: direction
      dtype: string
    - name: source
      dtype: string
    - name: model_prediction
      dtype: string
---

# UBC-NLP/AlexandriaX_Subtask_3_Test

This dataset contains the test split for AlexandriaX Subtask 3, a dialectal Arabic MT evaluation task.

Participants receive machine-translated outputs with source information. The goal is to detect and classify translation errors using LQM-inspired annotations.

Each row contains:

- `id`
- `direction`
- `source`
- `model_prediction`

## Example

```json
{
  "id": "43571_1",
  "direction": "ENG_UAE",
  "source": "Come on, be with her. It is nonsense. You do not know if she's moaning or not, and if she's moaning, it's called nocturnal moaning. We still too early.",
  "model_prediction": "يلا خلّك وياها، هذا خرابيط. انت ما تدري إذا هي تأن ولا لا، وإذا هي تأن يسمونه أنين ليلي. نحنا بعدنا وايد بدري."
}
```

## Load with `datasets`

```python
from datasets import load_dataset

dataset = load_dataset("json", data_files={"test": "test.jsonl"})
print(dataset["test"][0])
```