Skip to content

Release preview · native review pending

A small dataset built like areal lesson.

The MeloLingua Multilingual Graded Stories Dataset is a structured collection of newly written A1–B1 scenes. Each record connects aligned translation, contextual vocabulary, comprehension, grammar, and sentence practice—without exposing private, personalized, premium, audio, or book content.

Explore one record Read the method

36

original story records

12

target languages

A1–B1

balanced CEFR range

3

machine-readable files

Interactive record / ES-A2

Read first. Reveal only when meaning breaks.

Open a sentence to compare the English. The data preserves that same one-to-one alignment for every record.

Scene
Travel and directions
Outcome
Check a platform and correct a travel mistake politely.
Grammar
ir a + infinitive · polite questions · present tense

Spanish · A2

The Wrong Platform

Tap a line to reveal English

01Iker llega a la estación de Bilbao con diez minutos de margen.

Iker reaches Bilbao station with ten minutes to spare.

02Mira el panel, pero no encuentra el número de su tren.

He checks the board but cannot find his train number.

03Se acerca a una empleada y pregunta: «¿De qué andén sale el tren a Gernika?»

He walks over to an employee and asks, “Which platform does the train to Gernika leave from?”

04Ella observa su billete y responde: «Me parece que este tren sale de otra estación».

She looks at his ticket and replies, “I think this train leaves from another station.”

Vocabulary anchors

  • con diez minutos de margenwith ten minutes to spare · s1
  • andénplatform · s3
  • me parece queI think that / it seems to me · s4

Comprehension check

What is wrong with Iker’s travel plan?

Reveal answer +

His train leaves from another station.

The employee checks the ticket and explains that the train departs elsewhere.

Inside each record

Pedagogy is part of the data.

The schema records not only what a story says, but how a teacher or learning tool can use it. Every annotation points back to the scene instead of becoming a detached word list.

01

Aligned story text

Target-language sentences paired one-to-one with natural English translations.

02

Vocabulary in context

Useful words and chunks linked back to the exact sentence where they appear.

03

Reading checks

Recall and inference questions with choices, evidence references, and explanations.

04

Sentence building

Reconstruction exercises generated from real lines in each scene.

05

Teaching metadata

CEFR level, grammar focus, theme, locale, register, and communicative outcome.

06

Traceable provenance

Explicit public-sample origin, review state, and semantic hashes for version control.

Machine-readable by design

One source. Three useful views.

01

stories.json

Nested canonical records for applications, research, and complete pedagogy.

02

stories.csv

One row per story with nested learning fields serialized as JSON.

03

sentences.csv

One row per target-language and English sentence pair.

story.schema.json
{
  "id": "mls-es-a2-001",
  "language": { "code": "es", "locale": "es-ES" },
  "cefr_level": "A2",
  "theme": "travel and directions",
  "sentences": [
    {
      "id": "s3",
      "text": "¿De qué andén sale el tren a Gernika?",
      "translation": "Which platform does the train to Gernika leave from?"
    }
  ],
  "provenance": {
    "origin": "original-public-sample",
    "contains_private_content": false,
    "derived_from_book": false
  }
}

Release chain

Reviewed once. Citable everywhere.

The public release will use the same version, license, provenance statement, and canonical URL across every repository.

  1. 01

    GitHub

    Source, validation, changelog, and citation file

  2. 02

    Hugging Face

    Searchable language, task, license, and data cards

  3. 03

    Zenodo

    Immutable archive and DOI for formal citation

  4. 04

    Kaggle + educator directories

    Additional developer and classroom discovery

Balanced language matrix

Twelve languages. Three levels each.

Version 1 is intentionally regular: one original story for every language and CEFR pair. That makes gaps obvious and review progress measurable.

  • Arabic
  • Dutch
  • French
  • German
  • Italian
  • Japanese
  • Korean
  • Polish
  • Portuguese
  • Russian
  • Spanish
  • Turkish

Questions before release

Clear boundaries make useful data.

Does this dataset contain MeloLingua’s private or paid stories?

No. Every record is written specifically for the public sample dataset. It excludes private and personalized stories, book manuscripts, premium catalogue exports, audio, and learner data.

Is the dataset a CEFR benchmark?

No. CEFR labels are transparent editorial classifications for teaching use. The first release is intentionally small and is not a statistically representative or psychometrically validated benchmark.

Can teachers adapt the stories?

Yes, for non-commercial use under CC BY-NC 4.0. Attribution is required, and adaptations should clearly state that changes were made.

When will downloads and a DOI be available?

After every release record passes qualified native review. The approved files will then be released through GitHub and Hugging Face and archived on Zenodo for a DOI.

Public release gate

No story ships on structure alone.

All 36 records are structurally validated drafts. Downloads, repository visibility, and the DOI remain closed until qualified native review is complete for the exact semantic hashes.

Read MeloLingua’s editorial policy →