Parallel Literary Corpora: Fiction and Poetry Translations

This repository provides high-quality parallel datasets for translating fiction and poetry between Russian and various foreign languages.

Key Features

• Paragraph-level alignment: Unlike typical parallel corpora, this dataset avoids sentence-level alignment to better reflect literary translation practices. In fiction and poetry, translators often work with larger textual units (paragraphs, stanzas, or passages) rather than individual sentences.
• Diverse languages: Covers multiple languages with emphasis on English and German translations.

Dataset Structure

Current Version: Fiction Translations

File: prose_translation_pairs.public.jsonl.gz
Samples: 42,621
Format: gzip'ed JSON lines

Each entry contains:

• left_text: Source text (non-Russian literary passage)
• left_language: Source language name
• right_text: Russian translation
• domain: Always "prose" in current version

Data Statistics

Text Length Characteristics

Left text length (characters):

• Min: 51
• Max: 9,982
• Average: 408

Language Distribution

Language	Samples	Percentage
English	23,114	54.23%
German	19,370	45.45%
French	76	0.18%
Spanish	27	0.06%
Slovenian	12	0.03%
Catalan	10	0.02%
Polish	7	0.02%
Danish	2	<0.01%
Chinese	1	<0.01%
Persian	1	<0.01%
Dutch	1	<0.01%

Semantic Similarity Analysis

Text pairs were embedded using the multilingual-e5-large model. The histogram below shows the distribution of cosine similarity scores between source texts and their Russian translations:

Intended Use Cases

This dataset is designed to support research and development in the following areas:

1. Machine Translation (MT)

• Training and evaluating literary-aware MT systems that handle paragraph-level context
• Studying translation style transfer
• Developing domain-adapted models for fiction and poetry

2. Computational Creativity

• Analyzing creative translation choices (e.g., metaphors, cultural adaptations)
• Generating literary-aware paraphrases
• Studying cross-lingual stylistic patterns

3. Large Language Models

• Fine-tuning for literary text generation
• Evaluating cross-lingual coherence in long-form text
• Few-shot learning for low-resource literary languages

Future Additions

• Poetry translation pairs
• Expanded language coverage
• Additional metadata (author, period, genre)

Limitations

• Not suitable for sentence-level MT tasks without realignment
• Biased toward English/German translations (see Language Distribution)

Ethical Considerations

• Contains artistic works: Cite original authors where possible
• Verify licensing status of source texts before commercial use

Citation

If you use this dataset, please cite as:

@dataset{koziev_literary_translations_2024,
  author       = {Koziev},
  title        = {Parallel Literary Corpora: Russian Fiction Translations},
  year         = {2024},
  publisher    = {Zenodo},
  version      = {1.0},
  doi          = {10.5281/zenodo.15142403},
  url          = {https://github.com/Koziev/Translations},
  note         = {Paragraph-aligned parallel texts for literary translation between Russian and multiple languages},
  copyright    = {Creative Commons Attribution 4.0 International}
}