Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,33 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
## Latin part of cc100 corpus
|
| 2 |
+
This dataset contains parts of the Latin part of the [cc100](http://data.statmt.org/cc-100/) dataset. It was used to train a [RoBERTa-based LM model](https://huggingface.co/pstroe/roberta-base-latin-cased) with huggingface.
|
| 3 |
+
|
| 4 |
+
### Preprocessing
|
| 5 |
+
|
| 6 |
+
I undertook the following preprocessing steps:
|
| 7 |
+
|
| 8 |
+
- Removal of all "pseudo-Latin" text ("Lorem ipsum ...").
|
| 9 |
+
- Use of [CLTK](http://www.cltk.org) for sentence splitting and normalisation.
|
| 10 |
+
- Retaining only lines containing letters of the Latin alphabet, numerals, and certain punctuation (--> `grep -P '^[A-z0-9ÄÖÜäöüÆ挜ᵫĀāūōŌ.,;:?!\- Ęę]+$' la.nolorem.tok.txt`
|
| 11 |
+
- deduplication of the corpus
|
| 12 |
+
|
| 13 |
+
The result is a corpus of ~390 million tokens.
|
| 14 |
+
|
| 15 |
+
### Structure
|
| 16 |
+
The dataset is structured the following way:
|
| 17 |
+
```
|
| 18 |
+
{
|
| 19 |
+
"train": {
|
| 20 |
+
"text": "Solventibus autem illis pullum , dixerunt domini ejus ad illos : Quid solvitis pullum ?",
|
| 21 |
+
"text": "Errare humanum est ."
|
| 22 |
+
...
|
| 23 |
+
}
|
| 24 |
+
"test": {
|
| 25 |
+
"text": "Alia iacta est ."
|
| 26 |
+
...
|
| 27 |
+
}
|
| 28 |
+
}
|
| 29 |
+
```
|
| 30 |
+
|
| 31 |
+
### Contact
|
| 32 |
+
|
| 33 |
+
For contact, reach out to Phillip Ströbel [via mail](mailto:[email protected]) or [via Twitter](https://twitter.com/CLingophil).
|