---
language:
- uga
pretty_name: Copenhagen Ugaritic Corpus
license: mit
size_categories:
- 1K<n<10K
task_categories:
- fill-mask
- text-generation
tags:
- Ugaritic
- Semitic
- dead-language
- ancient-language
- ANE
- epigraphy
- cuneiform
configs:
- config_name: default
  data_files:
  - split: train
    path: data/cuc.parquet
dataset_info:
  config_name: default
  features:
  - name: row_id
    dtype: int64
  - name: tablet
    dtype: string
  - name: column
    dtype: string
  - name: line
    dtype: int64
  - name: ref
    dtype: string
  - name: text
    dtype: string
  - name: ugaritic_text
    dtype: string
  - name: text_html
    dtype: string
  - name: ugaritic_text_html
    dtype: string
  splits:
  - name: train
    num_bytes: 2763363
    num_examples: 7616
  download_size: 574013
  dataset_size: 2763363
---

# Copenhagen Ugaritic Corpus

A line-by-line corpus of Ugaritic alphabetic-cuneiform texts (Ras Shamra),
keyed to the **KTU** numbering (*Die keilalphabetischen Texte aus Ugarit*).
Each row is one text line, given in both Latin transliteration and Ugaritic
cuneiform, with markup distinguishing restored/damaged and uncertain readings.

- **279 tablets**, **7,616 lines**.
- Built from Copenhagen Ugaritic Corpus **v0.2.7** Text-Fabric data.
- Source project: [DT-UCPH/cuc](https://github.com/DT-UCPH/cuc)

## Usage

```python
from datasets import load_dataset

ds = load_dataset("AlexWalhai/cuc", split="train")
print(ds[0]["text"])           # transliteration
print(ds[0]["ugaritic_text"])  # cuneiform

# all lines of one tablet, in order
ktu_1_1 = ds.filter(lambda r: r["tablet"] == "KTU 1.1")
```

## Columns

| column               | type   | description                                            |
| -------------------- | ------ | ------------------------------------------------------ |
| `row_id`             | int64  | globally unique row index                              |
| `tablet`             | string | KTU tablet id, e.g. `KTU 1.1`                          |
| `column`             | string | column reference (Roman numeral), e.g. `II`            |
| `line`               | int64  | line number within the column                          |
| `ref`                | string | full reference, e.g. `KTU 1.1 II 1`                    |
| `text`               | string | Latin transliteration                                  |
| `ugaritic_text`      | string | Ugaritic cuneiform (Unicode)                           |
| `text_html`          | string | transliteration with bracket/uncertainty markup        |
| `ugaritic_text_html` | string | cuneiform with bracket/uncertainty markup              |

The HTML variants wrap restorations in `<span class="bracket">` and uncertain
sign readings in `<span class="uncertain">` for rendered display.

## Build

[`build_parquet.py`](build_parquet.py) builds `data/cuc.parquet` directly from
the upstream CUC Text-Fabric data (release `v0.2.7`) — no intermediate files.
Text-Fabric fetches the corpus to `~/text-fabric-data/github/DT-UCPH/cuc/tf/`:

```bash
python build_parquet.py            # uses ~/text-fabric-data/.../tf/0.2.7
python build_parquet.py /path/to/cuc/tf/0.2.7   # or an explicit TF data dir
```

## License

MIT. Underlying KTU references follow Dietrich, Loretz & Sanmartín,
*The Cuneiform Alphabetic Texts from Ugarit* (KTU³).