ECHOES · Cultural Heritage Cloud
Stage 1 Proposal · Bilingual recipe corpus
Three thousand recipes,
Three thousand recipes,
two languages,
one open corpus.
Romanian Academy · RACAI · USAMV · Casa Artusi
2026 · v1.0
Part one
The corpus
01
From manuscript
to machine.
SAVOR / Tavola
02 / 10
1.1 — Audiences
Who SAVOR is for
Three audiences,
Three audiences,
one shared corpus.
SAVOR is designed to be legible at three altitudes — to the curator who works folio by folio, to the researcher who works at scale, and to the cook who simply wants to know what's for dinner.
- Digital humanistsA clean, validated corpus structured against schema.org/Recipe and FOODon. Provenance preserved per record.
- NLP researchersMultilingual LaBSE embeddings, FAISS index, REST + SPARQL surfaces for cross-lingual retrieval.
- The curiousSunday-lunch dishes from Cluj alongside Sunday-lunch dishes from Romagna. What changes, what doesn't.
SAVOR / Tavola
03 / 10
1.2 — A recipe in two languages
RO · Romana
Plăcintă cu varză și ciuperci
IT · Italiano
Sfoglia ripiena di cavolo e funghi
a fi tinută la rece
o oră înainte — bunica
o oră înainte — bunica
SAVOR / Tavola
04 / 10
1.3 — By the numbers
The corpus, at a glance
3,300 recipes,
3,300 recipes,
three sources, two languages.
2,500
Romania
2,000 interwar manuscript recipes + 500 contemporary student recipes from USAMV Cluj.
800
Italia
Pellegrino Artusi's 1891 cookbook, digitised in XML by Casa Artusi. Primarily Emilia-Romagna.
181k
RDF triples
Linked-data graph in Turtle, ingested into the ECHOES Cultural Heritage Cloud.
SAVOR / Tavola
05 / 10
1.4 — On the value of structured heritage
A bilingual recipe corpus is not a database of dishes — it is the kitchen as archive, the archive as conversation.
Prof. Massimo Montanari
Food historian, Casa Artusi · formerly University of Bologna
SAVOR / Tavola
06 / 10
2.0 — Methodology
From manuscript to Zenodo
Six stages, twelve months,
Six stages, twelve months,
three quality gates.
01
Source consolidation
XML, CSV, OCR text harmonised in UTF-8 with multilingual titling.
M1 — M2
02
Schema & validation
SAVOR-JSON extends schema.org/Recipe. CI on every record.
M1 — M4
03
Semantic annotation
spaCy + FOODon + Wikidata. GeoNames + PeriodO for context.
M3 — M7
04
AI-readiness
LaBSE embeddings, FAISS, REST API, Docker container, notebooks.
M4 — M10
05
Quality assurance
Schema validation, linguistic proofreading, expert content review.
ongoing
06
Cloud ingestion
RDF/Turtle, BagIt, ECHOES PIDs, OAI-PMH, Zenodo DOI.
M8 — M12
SAVOR / Tavola
07 / 10
3.1 — SAVOR-JSON · the schema
Lab register · the dataset
Every recipe
Every recipe
is a record.
SAVOR-JSON extends schema.org/Recipe with culinary-heritage-specific fields. The metadata spine is Qualified Dublin Core for ECHOES Cloud interoperability; recipe detail is carried in schema.org and FOODon. Every record validates before ingestion.
// /recipes/ro/cluj/1934/placinta-varza-ciuperci.json { "@type": "Recipe", "recipeId": "sav:ro/cluj/1934/placinta-varza-ciuperci", "name": { "ro": "Plăcintă cu varză și ciuperci", "it": "Sfoglia ripiena di cavolo e funghi" }, "region": "geonames:665087", "period": "periodo:p0qhvzd", "festive": ["wd:Q3406641"], "diet": ["vegan", "de-post"], "license": "CC-BY-SA-4.0", "embedding": "<float[768], LaBSE>" }
SAVOR / Lab
08 / 10
3.2 — Architecture
Pipeline
Source → Schema → Semantic → Surface.
01 / SOURCE
Raw text
- XML · Casa Artusi
- CSV · USAMV student
- OCR · RO archives
02 / SCHEMA
SAVOR-JSON
- schema.org/Recipe
- Dublin Core spine
- CI validation
03 / SEMANTIC
Annotation
- FOODon · Wikidata
- GeoNames · PeriodO
- FAO · nutrition
04 / SURFACE
Access layer
- REST · /v1/converse
- SPARQL endpoint
- FAISS · LaBSE
→→→
SAVOR / Lab
09 / 10
Thank you
SAVOR · 2026
A bilingual
A bilingual
corpus, made open.
RO · Coordinator
Romanian Academy
Cluj-Napoca
Cluj-Napoca
RO
RACAI
"Mihai Drăgănescu"
"Mihai Drăgănescu"
RO
USAMV
Cluj-Napoca
Cluj-Napoca
IT
Casa Artusi
Forlimpopoli
Forlimpopoli
savor.eu
github.com/savor-corpus
doi:10.5281/zenodo.8729471
ECHOES · Cultural Heritage Cloud
10 / 10