SINAI/ALIA-es-cultural

Dataset Introduction

The ALIA Spanish Cultural and Heritage Corpus is a strategic open data infrastructure designed to support research and innovation in digital humanities, cultural analytics, and Spanish-language NLP. It consolidates heterogeneous official and academic repositories into a single curated dataset, enabling broad and structured access to cultural heritage documentation from Spain. With 236,399 instances, 946,467,028 tokens and 102 source datasets, it provides a robust foundation for developing domain-specialized language technologies in the cultural heritage field.

Dataset Details

Dataset Description

The ALIA Spanish Cultural and Heritage Corpus is an open-access data resource that compiles and organizes a large-scale collection of cultural heritage documents in Spanish. It integrates heritage inventories, specialized journals, archival records, institutional publications, and descriptive resources about tangible and intangible heritage.

The corpus was designed to provide a homogeneous and reusable textual base for researchers in digital humanities, cultural institutions, archivists, historians, linguists, and AI practitioners. Its breadth supports both documentary exploration and computational workflows such as semantic retrieval, topic discovery, terminology extraction, and language model adaptation.

The processed version of the corpus currently includes 236,399 instances and 946,467,028 tokens, distributed across 102 source datasets. This scale and heterogeneity make it an important resource for building and evaluating NLP systems focused on cultural heritage narratives, historical discourse, and institutional documentation in Spanish.

Uses

The primary purpose of this corpus is to support training and evaluation of language technologies specialized in Spanish cultural heritage, including:

• Training large language models (LLMs) specialized in cultural heritage text.
• Heritage-oriented information retrieval systems.
• Question-answering systems over cultural and historical documentation.
• NLP research in digital humanities and institutional heritage archives.

Dataset Structure

Data Instances

Each instance in the corpus follows this structure:

{
  "id": "19467",
  "text": "El 'Castillo de Jaén' es, en realidad, un conjunto de tres castillos o fortalezas que conforman un gran recinto, que ocupa casi toda la alargada cima del Cerro de Santa Catalina, situado, en su parte sur-occidental, junto a la ciudad española de Jaén, comunidad autónoma de Andalucía...",
  "source_id": "Wikipedia_Cultura_España",
}

Data Fields

• id (string): Document identifier extracted from the source (in many sources it corresponds to the document title or record identifier).
• text (string): Cleaned textual content of the document.
• source_id (string): Source dataset of origin.

Data Splits

The complete dataset contains the following source datasets with their statistics, sorted by number of tokens:

Dataset	Num Tokens	Num Instances	Tokens Percentage	Instances Percentage
Revistas_Culturales_Biblioteca_Virtual_Prensa_Historica	262,539,044	3,079	27.7388%	1.3025%
Wikipedia_Cultura_España	109,529,621	92,044	11.5725%	38.9359%
Revistas_CSIC	80,141,134	5,086	8.4674%	2.1514%
Revista_Hidalguia	34,552,535	347	3.6507%	0.1468%
Revistas_Instituto_Andaluz_Patrimonio_Historico	26,463,170	4,427	2.7960%	1.8727%
Repositorio_Activos_Digitales_Andalucia	25,072,951	1,470	2.6491%	0.6218%
Revista_Folklore	20,532,402	408	2.1694%	0.1726%
Tabula	19,345,563	1,826	2.0440%	0.7724%
Revista_Memoria_Ecclesiae	16,257,174	81	1.7177%	0.0343%
Guia_Digital_Patrimonio_Andalucia_Patrimonio_Mueble	13,558,785	78,148	1.4326%	33.0577%
Revista_En_La_España_Medieval	13,095,998	742	1.3837%	0.3139%
Revista_Complutum	12,871,996	708	1.3600%	0.2995%
Revista_Signa	12,727,638	1,230	1.3448%	0.5203%
Revista_Baetica	12,691,369	1,030	1.3409%	0.4357%
Guia_Digital_Patrimonio_Andalucia_Patrimonio_Inmueble	12,156,307	18,287	1.2844%	7.7357%
Revista_Arte_Individuo_Y_Sociedad	12,009,920	1,168	1.2689%	0.4941%
Libros_Instituto_Geografico_Nacional	11,046,546	82	1.1671%	0.0347%
Revista_Hipogrifo	10,902,761	1,082	1.1519%	0.4577%
Revista_Cuadernos_De_Ilustracion_Y_Romanticismo	10,651,320	798	1.1254%	0.3376%
Revista_Cuadernos_De_Historia_Moderna	10,003,156	876	1.0569%	0.3706%
Revista_Ilu	8,293,223	773	0.8762%	0.3270%
Revista_Lucentum	8,063,553	389	0.8520%	0.1646%
Revista_Cuadernos_De_Historia_Contemporanea	8,029,086	834	0.8483%	0.3528%
Revistas_Cultura_Jaume_I	7,792,816	631	0.8234%	0.2669%
Publicaciones_Patrimonio_Subdireccion_Defensa	7,569,501	77	0.7998%	0.0326%
Revista_Saguntum	7,390,053	539	0.7808%	0.2280%
Patrimonio_Anales_De_Historia_Del_Arte	7,350,468	716	0.7766%	0.3029%
Revista_Cuadernos_De_Prehistoria_Y_Arqueologia	7,088,588	429	0.7490%	0.1815%
Revista_Investigaciones_De_Historia_Economica	7,078,046	814	0.7478%	0.3443%
Patrimonio_Arqueologia_Y_Territorio_Medieval	6,861,392	373	0.7249%	0.1578%
BOE_Patrimonio	6,703,747	8	0.7083%	0.0034%
Revista_El_Futuro_Del_Pasado	6,247,979	14	0.6601%	0.0059%
Revista_Investigaciones_Historicas_Epoca_Moderna_Y_Contemporanea	5,578,182	478	0.5894%	0.2022%
Patrimonio_Boletin_De_Literatura_Oral	5,328,499	266	0.5630%	0.1125%
Tesis_Palacios	5,141,310	10	0.5432%	0.0042%
Revista_Castilla	5,020,526	578	0.5304%	0.2445%
Revista_Studia_Aurea	5,005,694	335	0.5289%	0.1417%
Guia_Digital_Patrimonio_Andalucia_Patrimonio_Inmaterial	4,902,654	1,845	0.5180%	0.7805%
Revista_Estudios_Romanicos	4,806,804	515	0.5079%	0.2179%
Revista_Pasavento	4,594,416	453	0.4854%	0.1916%
Revista_Paleohispanica	4,331,377	333	0.4576%	0.1409%
Revista_Acotaciones	4,261,752	504	0.4503%	0.2132%
Revista_Cauriensia	4,079,146	367	0.4310%	0.1552%
Revista_Anuario_Lope_De_Vega	3,967,112	409	0.4192%	0.1730%
Revista_Panta_Rei	3,606,096	281	0.3810%	0.1189%
Revista_Janus	3,486,313	218	0.3684%	0.0922%
Revista_Rubrica_Contemporanea	3,380,441	232	0.3572%	0.0981%
Revista_Quiroga	3,252,848	379	0.3437%	0.1603%
Revista_Brumal	3,191,857	181	0.3372%	0.0766%
Revista_Edad_Media	2,908,244	319	0.3073%	0.1349%
Revista_Saitabi	2,760,118	189	0.2916%	0.0799%
Revista_Anales_De_Arqueologia_Cordobesa	2,741,395	170	0.2896%	0.0719%
Revista_Anuario_Calderoniano	2,477,230	267	0.2617%	0.1129%
Publicaciones_Patrimonio_Cultural_Madrid	2,464,665	43	0.2604%	0.0182%
Revista_Edad_De_Oro	2,323,803	237	0.2455%	0.1003%
Revista_Aragon_En_La_Edad_Media	2,282,252	120	0.2411%	0.0508%
Revista_Santander_Estudios_Patrimonio	2,151,563	134	0.2273%	0.0567%
Revista_Molinum	2,150,286	24	0.2272%	0.0102%
Revista_Electronica_Complutense_De_Investigacion_En_Educacion_Musical	2,129,571	158	0.2250%	0.0668%
Revista_Escritura_Imagen	2,007,655	218	0.2121%	0.0922%
Patrimonio_Cuadernos_De_Arte_Prehistorico	1,934,449	127	0.2044%	0.0537%
Revista_MuseosEs	1,737,966	8	0.1836%	0.0034%
Revista_Potestas	1,668,484	129	0.1763%	0.0546%
Revista_Eikon	1,639,156	136	0.1732%	0.0575%
Revista_UcoArte	1,638,460	140	0.1731%	0.0592%
Revista_De_Medio_Aevo	1,506,558	115	0.1592%	0.0486%
Revista_Imago	1,437,216	175	0.1519%	0.0740%
Revista_Pygmalion	1,385,374	230	0.1464%	0.0973%
Revista_Historia_Social_Y_De_La_Educacion	1,381,186	113	0.1459%	0.0478%
Revista_Amaltea	1,260,417	126	0.1332%	0.0533%
Patrimonio_Castilla_Y_Leon	1,189,827	1,049	0.1257%	0.4437%
Revista_ReVisiones	1,172,930	150	0.1239%	0.0635%
Actas_De_Arquitectura_Religiosa_Contemporanea	1,147,371	145	0.1212%	0.0613%
Revista_Estudis	1,127,559	105	0.1191%	0.0444%
Revista_ASRI	1,118,663	102	0.1182%	0.0431%
Revista_El_Pajaro_De_Benin	929,162	69	0.0982%	0.0292%
Revista_Ad_Limina	878,163	7	0.0928%	0.0030%
Revista_Andelma	858,942	168	0.0908%	0.0711%
Revista_AusArt	841,225	107	0.0889%	0.0453%
Revista_Otarq	797,772	4	0.0843%	0.0017%
Mineralogia_Topologia_Iberica_Lamparas	712,411	6	0.0753%	0.0025%
Mineralogia_Topologia_Iberica_Acopios	664,062	13	0.0702%	0.0055%
Mineralogia_Topologia_Iberica_Hastial	654,350	10	0.0691%	0.0042%
Revista_Sarmental	538,358	42	0.0569%	0.0178%
Ministerio_De_Cultura_Patrimonio_Filmoteca_Española	506,171	3,177	0.0535%	1.3439%
Bienes_Culturales_Castilla_LaMancha	440,081	628	0.0465%	0.2657%
Revista_Riparia	406,607	44	0.0430%	0.0186%
Mineralogia_Topologia_Iberica_Amalgama	318,606	45	0.0337%	0.0190%
Patrimonio_Cultural_Inmaterial_Comunidades_Autonomas	259,436	168	0.0274%	0.0711%
Revista_Buñueliana	241,062	27	0.0255%	0.0114%
Revista_Crater	210,400	20	0.0222%	0.0085%
Patrimonio_Cataluña	140,534	331	0.0148%	0.1400%
Obras_Singulares_Museos_Andalucia	140,268	368	0.0148%	0.1557%
Guia_Digital_Patrimonio_Andalucia_Paisaje_Cultural	109,668	116	0.0116%	0.0491%
Ministerio_De_Cultura_Patrimonio_Audiovisual_Cine_Español	25,115	167	0.0027%	0.0706%
UNESCO	17,011	70	0.0018%	0.0296%
Somos_Patrimonio	9,318	48	0.0010%	0.0203%
Adquisiciones_Archivo_Historico	9,236	25	0.0010%	0.0106%
Fiestas_Patrimoniales	8,755	19	0.0009%	0.0080%
Patrimonio_Cultural_Inmaterial_España	6,519	13	0.0007%	0.0055%
Patrimonio_Cultural_Inmaterial_UNESCO	652	1	0.0001%	0.0004%
TOTAL	946,467,028	236,399	100.0000%	100.0000%

Example Usage

To load the dataset:

from datasets import load_dataset

# Load the complete dataset
data = load_dataset("SINAI/ALIA-es-cultural", trust_remote_code=True)

# Load with streaming
stream_data = load_dataset("SINAI/ALIA-es-cultural", trust_remote_code=True, streaming=True)

Example of data access:

example = data["train"][0]
print(f"ID: {example["id"]}")
print(f"Source: {example["source_id"]}")
print(f"Tokens: {example["tokens"]}")
print(f"Text: {example["text"][:200]}...")

Dataset Creation

Curation Rationale

This corpus was created to address the need for large, structured, and legally reusable linguistic resources in the cultural heritage domain in Spanish. It supports the ALIA initiative by providing a curated documentary foundation for training AI models and conducting computational analysis of cultural assets, historical publications, and institutional heritage records.

Source Data

The corpus integrates documentation from multiple publicly accessible repositories, including:

Heritage Inventories and Catalogs

• Regional and national heritage inventories (e.g., Castilla y León, Castilla-La Mancha, Cataluña).
• Digital guides of Andalusian property heritage (movable, immovable, intangible heritage, and cultural landscapes).
• Intangible cultural heritage portals (national, autonomous communities, and UNESCO-oriented records).

Institutional and Public Heritage Repositories

• Instituto Andaluz del Patrimonio Histórico (IAPH) repositories and publications.
• Public digital publications on heritage and culture (including Ministry of Defense heritage publications).
• Specialized resources such as archival works and acquisitions.

Academic Journals and Humanities Publications

• Broad collections of peer-reviewed journals and historical/cultural periodicals (e.g., Revistas CSIC, Biblioteca Virtual de Prensa Histórica cultural magazines, archaeology and history journals).
• Thematic journals covering art history, archaeology, medieval studies, folklore, philology, cultural studies, and conservation.

Audiovisual and Documentary Heritage

• Spanish audiovisual heritage records (ICAA catalog resources and Filmoteca Española-related material).
• Documentary and bibliographic heritage references from institutional collections.

Open Knowledge and Reference Resources

• Cultural heritage content from Spanish Wikipedia and complementary reference repositories.
• UNESCO-related and dissemination-oriented heritage resources.

All data come from official, institutional, academic, or publicly accessible sources.

Data Collection and Processing

The current corpus version was processed with an advanced documento parsing tool MinerU and a cleaning methodology based on datatrove. The processing pipeline includes the following steps:

Step 1: Source consolidation and normalization

• Integration of heterogeneous source formats into a unified schema.
• Normalization of identifiers and source tags.

Step 2: Language and quality filtering

• Selection and preservation of Spanish textual content.
• Basic quality checks to remove malformed records.

Step 3: Deduplication and cleaning

• Reduction of repeated or highly similar content.
• Text cleaning and standardization for downstream NLP usage.

Step 4: Quality filters and final cleaning

• Application of multiple specialized filters
• Correction of encoding errors
• Removal of sensitive or identifiable information

Token counting was performed using tiktoken.

Final corpus size in the latest parquet release:

• Tokens: 946,467,028
• Instances: 236,399
• Sources: 102

Annotations

The dataset does not provide task-specific manual annotations. It contains structural metadata fields generated during collection and preprocessing.

Personal and Sensitive Information

The corpus is assembled from public sources and has undergone processing steps oriented to large-scale textual reuse. Some documents may still contain personal names or role references appearing in public institutional or academic official contexts. Users are advised to apply additional controls depending on the specific use of the corpus.

Considerations for Using the Data

Social Impact of Dataset

This corpus contributes to democratizing access to cultural heritage documentation in Spanish. It can support AI systems for heritage dissemination, documentation support, educational access, and research acceleration in humanities and social sciences.

Discussion of Biases

Potential biases include:

• Institutional register bias: The corpus is dominated by formal and academic language, not colloquial speech.
• Source concentration bias: A reduced number of high-volume sources contributes a substantial share of total tokens.
• Regional representation bias: Coverage depends on digitization and publication practices across institutions and territories.
• Historical and editorial bias: Heritage narratives may reflect dominant historiographic, institutional, or curatorial perspectives.

Other Known Limitations

• Source OCR and digitization quality may vary across collections.
• Temporal coverage is uneven across source datasets.
• Some records include highly structured or catalog-like text that may differ from narrative prose.
• Token and source statistics are specific to the current enriched parquet release and may change in future updates.

Citation

@misc{ALIA-es-cultural,
  title={ALIA Spanish Cultural and Heritage Corpus},
  author={SINAI Research Group},
  year={2026},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/datasets/SINAI/ALIA-es-cultural}}
}

---

Funding

This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ALIA.

Acknowledgments

This dataset has been generated thanks to CEATIC (Centro de Estudios Avanzados en Tecnologías de la Información y de la Comunicación) – UJA (Universidad de Jaén) which provided the needed computational resources on its clusters.

---

Contact: ALIA Project - SINAI Research Group - Universidad de Jaén

More Information: SINAI Research Group | ALIA-UJA Project