Datasets:
SQAC

Task Categories: question-answering
Languages: es
Multilinguality: monolingual
Size Categories: unknown
Language Creators: found
Annotations Creators: expert-generated
Source Datasets: original

SQAC (Spanish Question-Answering Corpus): An extractive QA dataset for the Spanish language

BibTeX citation

@article{DBLP:journals/corr/abs-2107-07253,
  author    = {Asier Guti{\'{e}}rrez{-}Fandi{\~{n}}o and
               Jordi Armengol{-}Estap{\'{e}} and
               Marc P{\`{a}}mies and
               Joan Llop{-}Palao and
               Joaqu{\'{\i}}n Silveira{-}Ocampo and
               Casimiro Pio Carrino and
               Aitor Gonzalez{-}Agirre and
               Carme Armentano{-}Oller and
               Carlos Rodr{\'{\i}}guez Penagos and
               Marta Villegas},
  title     = {Spanish Language Models},
  journal   = {CoRR},
  volume    = {abs/2107.07253},
  year      = {2021},
  url       = {https://arxiv.org/abs/2107.07253},
  archivePrefix = {arXiv},
  eprint    = {2107.07253},
  timestamp = {Wed, 21 Jul 2021 15:55:35 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2107-07253.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

See the pre-print version of our paper for further details: https://arxiv.org/abs/2107.07253

Introduction

This dataset contains 6,247 contexts and 18,817 questions with their answers, 1 to 5 for each fragment.

The sources of the contexts are:

This dataset can be used to build extractive-QA.

Supported Tasks and Leaderboards

Extractive-QA

Languages

ES - Spanish

Directory structure

  • README.md
  • dev.json
  • test.json
  • train.json
  • sqac.py

Dataset Structure

Data Instances

JSON files

Data Fields

Follows (Rajpurkar, Pranav et al., 2016) for squad v1 datasets. (see below for full reference). We added a field "source" with the source of the context.

Example

{
  "data": [
    {
      "paragraphs": [
        {
          "context": "Al cogote, y fumando como una cafetera. Ah!, no era él, éramos todos nosotros. Luego llegó Billie Holiday. Bajo el epígrafe Arte, la noche temática, pasaron la vida de la única cantante del universo que no es su voz, sino su alma lo que se escucha cuando interpreta. Gata golpeada por el mundo, pateada, violada, enganchada a todos los paraísos artificiales del planeta, jamás encontró el Edén. El Edén lo encontramos nosotros cuando, al concluir la sesión de la tele, pusimos en la doméstica cadena de sonido el mítico Last Recording, su última grabación (marzo de 1959), con la orquesta de Ray Ellis y el piano de Hank Jones. Se estaba muriendo Lady Day, y no obstante, mientras moría, su alma cantaba, Baby, won't you please come home. O sea, niño, criatura, amor, vuelve, a casa por favor.",
          "qas": [
            {
              "question": "¿Quién se incorporó a la reunión más adelante?",
              "id": "c5429572-64b8-4c5d-9553-826f867b07be",
              "answers": [
                {
                  "answer_start": 91,
                  "text": "Billie Holiday"
                }
              ]
            },
            
            ...
            
            ]
        }
      ],
      "title": "P_129_20010702_&_P_154_20010102_&_P_108_20000301_c_&_P_108_20000601_d",
      "source": "ancora"
    },
    ...
  ]
}

Data Splits

  • train
  • development
  • test

Content analysis

Number of articles, paragraphs and questions

  • Number of articles: 3,834
  • Number of contexts: 6,247
  • Number of questions: 18,817
  • Questions/context: 3.01
  • Number of sentences: 48,026
  • Sentences/context: 7.70

Number of tokens

  • Total tokens in context: 1,561,616
  • Tokens/context 250.30
  • Total tokens in questions: 203,235
  • Tokens in questions/questions: 10.80
  • Tokens in questions/tokens in context: 0.13
  • Total tokens in answers: 90,307
  • Tokens in answers/answers: 4.80
  • Tokens in answers/tokens in context: 0.06

Lexical variation

46.38 of the words in the Question can be found in the Context.

Question type

Question Count %
qué 6,381 33.91 %
quién/es 2,952 15.69 %
cuál/es 2,034 10.81 %
cómo 1,949 10.36 %
dónde 1,856 9.86 %
cuándo 1,639 8.71 %
cuánto 1,311 6.97 %
cuántos 495 2.63 %
adónde 100 0.53 %
cuánta 49 0.26 %
no question mark 43 0.23 %
cuántas 19 0.10 %

Dataset Creation

Methodology

6,247 contexts were randomly chosen from the three corpus described below. We commisioned the creation of between 1 and 5 questions for each context, following an adaptation of the guidelines from SQUAD 1.0 Rajpurkar, Pranav et al. “SQuAD: 100, 000+ Questions for Machine Comprehension of Text.” EMNLP (2016). In total, 18,817 pairs of a question and an extracted fragment that contains the answer were created.

Curation Rationale

For compatibility with similar datasets in other languages, we followed as close as possible existing curation guidelines. We also created another QA dataset with Wikipedia to ensure thematic and stylistic variety.

Source Data

Initial Data Collection and Normalization

The source data are scraped articles from the Spanish Wikipedia site, Wikinews site and from AnCora corpus.

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

We commissioned the creation of 1 to 5 questions for each context, following an adaptation of the guidelines from SQUAD 1.0 Rajpurkar, Pranav et al. “SQuAD: 100, 000+ Questions for Machine Comprehension of Text.” EMNLP (2016).

Who are the annotators?

Native language speakers.

Dataset Curators

Carlos Rodríguez and Carme Armentano, from BSC-CNS.

Personal and Sensitive Information

No personal or sensitive information included.

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Contact

Carlos Rodríguez-Penagos (carlos.rodriguez1@bsc.es) and Carme Armentano-Oller (carme.armentano@bsc.es)

Funding

This work was partially funded by the Spanish State Secretariat for Digitalization and Artificial Intelligence (SEDIA) within the framework of the Plan-TL.

License

Attribution-ShareAlike 4.0 International License
This work is licensed under a Attribution-ShareAlike 4.0 International License.

Models trained or fine-tuned on BSC-TeMU/SQAC