---
license: cc-by-4.0
language:
- de
- fr
- it
- en
dataset_info:
  - config_name: de
  - config_name: it
  - config_name: fr
---

[![Paper](https://img.shields.io/badge/📄%20Paper-arXiv%3A2512.07538-B31B1B.svg)](https://arxiv.org/pdf/2512.07538)
[![GitHub Repo](https://img.shields.io/badge/GitHub-Repo-black?logo=github)](https://github.com/ZurichNLP/SwissGov-RSD)

# SwissGov-RSD

## Dataset Description

SwissGov-RSD is the a naturalistic, human-annotated, document-level, cross-lingual dataset for token-level semantic difference recognition (RSD). It contains 224 multi-parallel Swiss government documents from [admin.ch](admin.ch) in English–German, English–French, and English–Italian, annotated with fine-grained semantic difference labels (0–1) at the token level.
The dataset targets real-world scenarios where cross-lingual content diverges due to translation errors, asynchronous updates, or explicitations.

## Dataset Structure

The dataset consists of 224 document pairs for each language pair. Each sample consists of:
- text_a: English side of the document pair as a string.
- text_b: Non-English side of the document pair as a string.
- labels_a: Per word (separated by whitespaces) label for English side ranging from 0 to 1 (punctuation = -1).
- labels_b: Per word (separated by whitespaces) label for non-English side ranging from 0 to 1 (punctuation = -1).
- page_en: URL of the scraped webiste in the English language.
- page_other: URL of the scraped website in the non-English language.
- subset: Subset name according to language.
- id: The sample ID.


## Citation information

```bibtex
@misc{wastl2025swissgovrsdhumanannotatedcrosslingualbenchmark,
      title={SwissGov-RSD: A Human-annotated, Cross-lingual Benchmark for Token-level Recognition of Semantic Differences Between Related Documents}, 
      author={Michelle Wastl and Jannis Vamvas and Rico Sennrich},
      year={2025},
      eprint={2512.07538},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2512.07538}, 
}
```