ruscorpora/rnc_deru_aligned

RNC DeRu Aligned (Russian-German Word Alignment Dataset)

Dataset Description

This dataset contains sentence pairs with manual word alignment annotations. It is built upon the Russian-German subcorpus of the Russian National Corpus (RNC).

The primary purpose of this dataset is to support word alignment tasks. The annotation distinguishes between "Sure" and "Possible" alignment links, following standard alignment conventions.

• Curated by: Russian National Corpus
• Language(s): German, Russian
• License: CC BY-NC-SA 4.0

Dataset Statistics

• Current Size: 475 sentence pairs.
• Status: Active (The dataset is currently being expanded).

Dataset Structure

The dataset is organized into three directories representing the full set and the training/validation splits.

Directory Layout

• all/: Contains the complete dataset (all available annotated pairs).
• train/: Contains 80% of the data from all, intended for training.
• dev/: Contains the remaining 20% of the data, intended for development/validation.

File Format

Each directory contains three parallel files sharing the prefix deru:

1. deru.src: Source sentences in German. One sentence per line.
2. deru.tgt: Target sentences in Russian. One sentence per line.
3. deru.talp: Word alignment annotations in Pharaoh format.

The .talp files contain the alignment links for the corresponding sentence pairs.

• Notation: Space-separated pairs of indices i-j (e.g., 1-1 2p3).
• Indexing: 1-based indexing (The first word in the sentence has index 1, not 0).
• Link Types: The dataset contains both S (Sure) and P (Possible) links.

Dataset Creation

Annotation Process

Two native Russian speakers (German level B2+) performed manual annotation using Sure (S) and Possible (P) labels, based on principles from [Politova et al., 2023]. Language-specific Annotation Rules:

• Articles: Not aligned.
• Separable verb elements (reflexive particles & separable prefixes): Aligned together with the main word.
• Auxiliaries & Prepositions: Aligned as S if an explicit counterpart exists; otherwise marked as P (attached to the main word) or left unaligned.
• Idioms: Marked as S for stable translations, otherwise P.

Quality Control

Annotations underwent selective joint verification by all annotators to ensure consistency, though parallel double-annotation was not performed.

License

This dataset is distributed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

How to cite

If you are using this dataset in your scientific work, please, cite this paper:

TBA