jesterb32's picture
Release es translation, prompt v1
e23728c verified
|
Raw History Blame Contribute Delete
3.13 kB
---
pretty_name: ELI-Why (Questions Only), Spanish translation
license: mit
language:
- es
- en
multilinguality: translation
source_datasets:
- INK-USC/eli-why-only-questions
task_categories:
- question-answering
size_categories: 10K<n<100K
language_creators:
- machine-generated
annotations_creators:
- machine-generated
configs:
- config_name: default
data_files:
- split: train
path: train.jsonl
---
# ELI-Why (Questions Only), Spanish
An LLM translation of [ELI-Why (Questions Only)](https://huggingface.co/datasets/INK-USC/eli-why-only-questions) into Spanish, made by Learning Equality for evaluating offline educational search. It has 13,392 "why" questions, each labelled with a domain and a discipline.
## Provenance
- Source: [INK-USC/eli-why-only-questions](https://huggingface.co/datasets/INK-USC/eli-why-only-questions) at revision `8a34e9d7fac1a87e0a5ff035dae85c22f8c3d55c`, split `train`.
- Model: `gpt-5.6-terra`, temperature 1, 20 questions per request.
- Prompt: prompt version 1, [es.py](https://github.com/learningequality/offline-ai-benchmarks/blob/3eec0291a1db3234b4240fe1f20cc7c30eddb33b/eli_why_translations/languages/es.py).
- The source questions were themselves generated with GPT-4 and curated by the ELI-Why authors.
## Review
A fluent reader judged a seeded sample of 100 rows (seed 0) of the model's translations, before any human corrections:
| Verdict | Rows | Meaning |
|---|---|---|
| ok | 90 | a faithful translation |
| minor | 10 | the meaning is right, but the wording is awkward |
| meaning | 0 | the translation changes what is asked |
The verdicts are in [review.csv](review.csv). No row's translation was supplied by a person.
## Alignment with the source
Row `i` translates source row `i`, so the dataset has the same rows in the same order. `Domain`, `Discipline`, and `Question_en` are copied unchanged from the source, and `id` is `eli-` followed by the 5-digit row index. A sampler that depends only on row order, `Discipline`, and a seed picks the same rows here as in the source.
## Columns
- `id`: `eli-` and the 0-based source row index, e.g. `eli-00042`.
- `Question`: the Spanish question.
- `Question_en`: the English source question.
- `Domain`, `Discipline`: the source labels.
- `human_corrected`: true when a person supplied `Question` instead of the model.
```json
{
"id": "eli-00000",
"Question": "¿Por qué el hielo flota en el agua?",
"Question_en": "Why does ice float in water?",
"Domain": "STEM",
"Discipline": "physics",
"human_corrected": false
}
```
## License
Released under the MIT license, like the source dataset. This is a derivative of ELI-Why (Questions Only) by Joshi et al. Please cite the ELI-Why paper ("ELI-Why: Evaluating the Pedagogical Utility of Language Model Explanations", ACL 2025 Findings):
```bibtex
@inproceedings{joshi2025eliwhy,
title={{ELI-Why}: Evaluating the Pedagogical Utility of Language Model Explanations},
author={Joshi, Brihi and He, Keyu and Ramnath, Sahana and Sabouri, Sadra and Zhou, Kaitlyn and Chattopadhyay, Souti and Swayamdipta, Swabha and Ren, Xiang},
year={2025}
}
```