Datasets:

ZurichNLP
/

mediomatix-raw

license: cc-by-nc-sa-4.0
language:
  - rm
configs:
  - config_name: rm-surmiran
    data_files:
      - split: train
        path: data/rm-surmiran/train.jsonl
      - split: validation
        path: data/rm-surmiran/validation.jsonl
      - split: test
        path: data/rm-surmiran/test.jsonl
  - config_name: rm-sursilv
    data_files:
      - split: train
        path: data/rm-sursilv/train.jsonl
      - split: validation
        path: data/rm-sursilv/validation.jsonl
      - split: test
        path: data/rm-sursilv/test.jsonl
      - split: no_surmiran
        path: data/rm-sursilv/no_surmiran.jsonl
  - config_name: rm-puter
    data_files:
      - split: train
        path: data/rm-puter/train.jsonl
      - split: validation
        path: data/rm-puter/validation.jsonl
      - split: test
        path: data/rm-puter/test.jsonl
      - split: no_surmiran
        path: data/rm-puter/no_surmiran.jsonl
  - config_name: rm-sutsilv
    data_files:
      - split: train
        path: data/rm-sutsilv/train.jsonl
      - split: validation
        path: data/rm-sutsilv/validation.jsonl
      - split: test
        path: data/rm-sutsilv/test.jsonl
      - split: no_surmiran
        path: data/rm-sutsilv/no_surmiran.jsonl
  - config_name: rm-vallader
    data_files:
      - split: train
        path: data/rm-vallader/train.jsonl
      - split: validation
        path: data/rm-vallader/validation.jsonl
      - split: test
        path: data/rm-vallader/test.jsonl
      - split: no_surmiran
        path: data/rm-vallader/no_surmiran.jsonl

General Information

Release of the Mediomatix corpus, prepared by UZH and PHGR, to be used for research purposes only.

In mediomatix-raw, we release the full Mediomatix schoolbooks' text for each idiom. The data points are unaligned. See here for the mulit-parallel, aligned Mediomatix corpus.

We use the following Romansh idiom codes as subsets in the dataset:

Sursilvan: rm-sursilv
Sutsilvan: rm-sutsilv
Surmiran: rm-surmiran
Puter: rm-puter
Vallader: rm-vallader

The splits in each subset are:

train: Text from year 2 and 3 schoolbooks
validation: Text from year 4 schoolbooks
test: Text from year 5 schoolbooks
no_surmiran: Text from years 6-9; available for all idioms except rm-surmiran

Each data point contains the following columns:

CmsElementID: An id referring to which CMS element a text was extracted from.
segmentID: A unique id for each row; differentiates between segments split apart from the same CMS element.
text: The extracted plain text from within an HTML segment
html: The extracted text, including any HTML markup
chapter: The name of the chapter from which a text was extracted
book: A code referring to the grade, volume, and type of school book

Book names are encoded in the "book" column as follows:

The first digit refers to the school year of the book
The next refers to the book volums for that school year
The two letter code indicates whether the segment comes from the student workbook ("wb") or teacher's commentary ("tc")
Example: Book "5.1_tc" refers to teacher's commentary for Volume 1 of the year 5 schoolbook.

How to Use

The subset for a specific Romansh idiom must be specified when loading the datset:

from datasets import load_dataset

sursilv = load_dataset("ZurichNLP/mediomatix-raw","rm-sursilv")

# Load Train set:

train_set = sursilv['train']

# Load Validation set:

val_set = sursilv['validation']

# Load Test set:

test_set = sursilv['test']

# Load no-rm-surmiran if the subset is not rm-surmiran
no_surm = sursilv['no_surmiran']

Citation

@inproceedings{hopton-etal-2026-mediomatix,
    title = "The Mediomatix Corpus: Parallel Data for {R}omansh Language Varieties via Comparable Schoolbooks",
    author = {Hopton, Zachary William  and
      Vamvas, Jannis  and
      B{\"u}chler, Andrin  and
      Rutkiewicz, Anna  and
      Cathomas, Rico  and
      Sennrich, Rico},
    editor = "Demberg, Vera  and
      Inui, Kentaro  and
      Marquez, Llu{\'i}s",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EACL} 2026",
    month = mar,
    year = "2026",
    address = "Rabat, Morocco",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-eacl.16/",
    pages = "290--306",
    ISBN = "979-8-89176-386-9"
}