README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: int64
    - name: latin
      dtype: string
    - name: german
      dtype: string
    - name: source
      dtype: string
    - name: tag
      dtype: string
    - name: score
      dtype: float64
  splits:
    - name: train
      num_bytes: 154924781
      num_examples: 406011
  download_size: 91837604
  dataset_size: 154924781
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
task_categories:
  - translation
language:
  - la
  - de
size_categories:
  - 100K<n<1M
pretty_name: Latin-German-Textcorpus

📜 Latin-German Textcorpus

This dataset consists of 406,011 Latin-German parallel sentences (sentence pairs). Each entry contains a Latin sentence and its corresponding German translation. The sentence pairs were collected and processed from various websites and online sources.

📄 Dataset Schema

The dataset contains the following columns:

• id: A unique identifier for each entry.
• latin: The sentence in Latin.
• german: The German translation.
• source: The origin or reference from where the entry was taken.
• tag: A thematic category or label indicating the entry's historical period. Possible values are: 'ANCIENT', 'MEDIEVAL', 'MODERN', or 'UNKNOWN'.
• score: A numerical relevance or similarity score. A value of -1 indicates the score is undefined, while any value >= 1 represents a valid score.

📚 Citation / References

Falls Sie dieses Modell in Ihrer Forschung verwenden, bitten wir Sie, die zugrundeliegende Masterarbeit wie folgt zu zitieren:

Masterarbeit (Zenodo DOI):

Wenzel, M. (2025). Translatio ex Machina: Neuronale Maschinelle Übersetzung vom Lateinischen ins Deutsche [Zenodo]. Unveröffentlichte Masterarbeit, Fachhochschule Südwestfalen

DOI: 10.5281/zenodo.17940090

---

💻 Usage

Understanding the score Feature

The score column indicates the method and quality of the sentence alignment:

• score = -1: This value signifies that the Latin and German sentences were manually aligned or by a special tooling.
• score >= 1: This value indicates that the alignment was calculated by an automated alignment tool.
  • Interpretation: A higher score suggests a better alignment quality.
  • Recommendation: For high-confidence automated alignments, we recommend using only entries where the score is >= 1.2.

Loading and Filtering the Dataset

You can easily filter the dataset to select only high-quality alignments (manual alignments OR high-scoring automated alignments) using the filter() method:

from datasets import load_dataset, DatasetDict

# 1. Load the initial dataset (contains only the "train" split)
dataset = load_dataset("fhswf/latin-german-parallel")
train_dataset = dataset["train"]

# 2. Filter the dataset to include only high-quality alignments:
#    - Entries with score == -1 (manual alignment)
#    - Entries with score >= 1.2 (high-confidence automated alignment)
def filter_by_score(example):
    return example["score"] == -1 or example["score"] >= 1.2

high_quality_train = train_dataset.filter(filter_by_score)

# Optional: Proceed with splitting the high-quality data
temp_splits = high_quality_train.train_test_split(test_size=0.01, seed=42)

test_validation_splits = temp_splits["test"].train_test_split(test_size=0.5, seed=42)

dataset = DatasetDict({
    "train": temp_splits["train"],
    "validation": test_validation_splits["train"],
    "test": test_validation_splits["test"],
})