---
language:
  - en
license: cc-by-sa-4.0
configs:
- config_name: corpus
  data_files: "corpus.tsv"
  sep: "\t"
- config_name: queries
  data_files: "queries.tsv"
  sep: "\t"
- config_name: qrels
  data_files: "qrels/test.tsv"
  sep: "\t"
---

# Dataset Card for ApostolicFathersQA

### Dataset Summary

A dataset for evaluating retrieval from CCEL books.

### Supported Tasks and Leaderboards

[Evaluation Example](https://colab.research.google.com/drive/1oog-Hd2YL5qNdjrx40qgLgmZhypCIlOG)

## Dataset Structure

### Data Instances

This dataset includes 3 subsets:

#### Corpus

The corpus is 2.29k paragraphs from [The Apostolic Fathers with Justin Martyr and Irenaeus](https://ccel.org/ccel/schaff/anf01). The text of all `p` tags is included as long as the text greater than 160 characters. See the included `generate.ipynb` notebook for specific details of the corpus filtering.

Here's an example:

```json
{
  "_id": "ccel/s/schaff/anf01.xml:i.i-p1",
  "title": "",
  "text": "This volume, containing the equivalent of three volumes..."
}
```

The title field is always blank, but included to match the data format used by BIER and MTEB.

#### Queries

Queries are questions related to certain paragraphs in the corpus. They are formatted as follows:

```json
{
  "_id": "cain",
  "title": "",
  "text": "What did the apostolic fathers have to say about Cain and Abel?"
}
```

#### qrels

The final qrels subset links the corpus to queries. It is formatted as follows:

```json
{
  "query-id": "cain",
  "corpus-id":, "ccel/s/schaff/anf01.xml:v.ii.xii-p4",
  "score": 1
}
```

### Data Fields

- `_id`: A uniqie identifier matching the [CCEL paragraphs](https://huggingface.co/datasets/jncraton/ccel-paragraphs) dataset for corpus paragraphs.
- `title`: An empty string.
- `query-id`: Unique id of a query
- `corpus-id`: Unique id of a paragraph from the corpus
- `score`: A score for this pair

### Data Splits

There is currently only a single split. It is expected that this dataset will be primarily used for evaluation rather than training.

## Dataset Creation

### Source Data

#### Initial Data Collection and Normalization

The source data consists of freely licensed works from the CCEL.

#### Who are the source language producers?

This data has numerous authors. All are human. Dates of authorship vary and span several millenia.

### Personal and Sensitive Information

There should be no personally identifiable information for living humans in this dataset.

## Considerations for Using the Data

### Social Impact of Dataset

This dataset provides opportunities for ongoing research and exploration of classic Christian literature.

### Discussion of Biases

This dataset is biased towards classical Christian beliefs and perspectives.

## Additional Information

### Dataset Curators

The source dataset was curated and maintained by the numerous contributors to the CCEL.

### Licensing Information

[ApostolicFathersQA](https://huggingface.co/datasets/jncraton/ccel-paragraphs) is licensed under [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/).