Dataset Card for Letras Carnaval Cádiz

Name: letras-carnaval-cadiz
Creator: IES Rafael Alberti
License: https://choosealicense.com/licenses/cc-by-sa-4.0/

English | Español

Changelog

Release	Description
v1.0	Initial release of the dataset. Included more than 1K lyrics. It is necessary to verify the accuracy of the data, especially the subset midaccurate.

Dataset Summary

This dataset is a comprehensive collection of lyrics from the Carnaval de Cádiz, a significant cultural heritage of the city of Cádiz, Spain. Despite its cultural importance, there has been a lack of a structured database for these lyrics, hindering research and public access to this cultural heritage. This dataset aims to address this gap.

The dataset was created by the Cádiz AI Learning Community, a branch of the non-profit association Spain AI, and was developed by Iván Romero Reyna and Jesús Federico Franco Medinilla, students of the Specialization Course in Artificial Intelligence and Big Data at IES Rafael Alberti during the 2022-2023 academic year. The project is supervised by Jesús Carlos Avecilla de la Herrán, a computational linguist.

Collaboration is encouraged, with individuals able to verify the different records of the dataset at letrascarnavalcadiz.com, ensuring the transcription of the lyrics and all data are correct. New lyrics can also be added to the dataset. Corrections and additions are not immediately reflected in the dataset but are updated periodically.

For more information or to report a problem, you can write to [email protected].

Supported Tasks and Leaderboards

[More Information Needed]

Languages

The dataset is in Spanish, reflecting the language of the Carnaval de Cádiz.

Dataset Structure

Data Instances

A typical instance in the dataset is formatted in JSON and contains the following fields:

{
    "id": "9de8647521b728c45ff45c1c11208708d055397fd7781b31cf91b473dff224d5",
    "authors": ["Juan Carlos Aragón Becerra"],
    "song_type": 2,
    "year": "2018",
    "group": "Los Mafiosos",
    "group_type": 2,
    "lyrics": [
      "Mujer va llegando el momento",
      "de ser la que lleve la rienda",
      "el camino ha sido largo y polvoriento",
      "pero ya no habrá varón que te detenga",
      "gritad larga vida a la reina",
      "que va a comenzar tu gobierno",
      "ojalá no heredes nada",
      "de aquel macho que te odiaba",
      "porque en el fondo sabía",
      "que ya tú te le acercabas",
      "y el contigo no podía",
      "ten en cuenta cuando hagas justicia",
      "de volver a nivelar la balanza",
      "y aguantar aunque tragando saliva",
      "el deseo de venganza",
      "de ser oh humano fatal",
      "de ser o que puedo entender",
      "tan solo con una mirada",
      "la llaga que baña tu alma y tu piel",
      "que te sirva la experiencia",
      "del macho de la manada",
      "la fuerza no vale nada",
      "si no es con la inteligencia",
      "y ojalá que tu conciencia",
      "a mí me brinde la suerte",
      "de nunca volver a verte",
      "con los pies en una iglesia",
      "que ella fue quien escribió",
      "que ella fue quien escribió",
      "la historia contra vosotras",
      "y encima se la cobró",
      "y encima se la cobró",
      "con mil millones de devotas",
      "ojalá que tu corona y tu bandera",
      "abran paso a una vida nueva",
      "como un mundo en primavera",
      "ojalá que a ti no te envenene el poder",
      "y que no dejes nunca de ser la mujer",
      "que siempre fue nuestra gran compañera"
    ]
}

The id field uniquely identifies each instance in the dataset, providing a way to reference specific entries. The authors, song_type, year, group, and group_type fields provide context for the lyrics, while the lyrics field itself contains the actual text of the song. The relationships between these fields are implicit in the structure of the dataset, with each instance representing a single song from the Carnaval de Cádiz.

Data Fields

id

Unique identifier for each song in the dataset. A SHA-256 hash calculated from the first four verses of the lyrics and the group name, with all spaces removed and converted to lowercase (string).

authors

List of authors who have written the song (string array).

song_type

The type of song (1: presentación, 2: pasodoble/tango, 3: cuplé, 4: estribillo, 5: popurrí, 6: cuarteta).

year

Year the song was written or performed (string).

group

Name of the group that performed the song (string).

group_type

The type of the group (1: coro, 2: comparsa, 3: chirigota, 4: cuarteto).

lyrics

The lyrics of the song, represented as an array of verses (string array).

Data Splits

This dataset does not have traditional training, validation, and test splits. Instead, it is divided into two subsets: "accurate" and "midaccurate".

The "accurate" subset contains 958 instances. All fields of first 957 instances in this subset have been obtained through web scraping and have undergone at least one human review for accuracy. The rest have been added by users at letrascarnavalcadiz.com.

The "midaccurate" subset contains 226 instances. The 'group' and 'lyrics' fields in this subset were collected through web scraping, but the remaining fields were filled in by querying language models connected to the Internet. Therefore, the data in these fields may not be accurate.

Subset	Instances
Accurate	958
Midaccurate	226

Please note that the division into subsets is based on the method and reliability of data collection, rather than a random or stratified split typically used in machine learning tasks. Users of the dataset should consider this when deciding how to use the data.

Dataset Creation

Curation Rationale

The dataset was created to address a significant need in the cultural heritage of the city of Cádiz, Spain. The Carnaval de Cádiz is a major cultural event, yet there was no structured database of its lyrics that could be consulted for research or public access. This lack of a structured database hindered the exploration and appreciation of this cultural heritage. The dataset was curated to respond to this need.

Source Data

Initial Data Collection and Normalization

The initial collection of lyrics was carried out through automatic scraping of various websites and multimedia content on the Internet. To maximize the number of records with minimal effort, all collection is being done using different Artificial Intelligence models.

Who are the source language producers?

The source language producers of the dataset are the authors and performers of the songs from the Carnaval de Cádiz. These include a wide range of individuals and groups who have participated in the Carnaval over the years. The dataset does not include self-reported demographic or identity information for these individuals or groups.

The data in the dataset was collected from two websites: https://www.alsondelcarnaval.es and http://letrasdesdeelparaiso.blogspot.com. The first 957 instances of "accurate" subset of the dataset was collected from the former, while the "midaccurate" subset was collected from the latter. The data was extracted through automatic web scraping, and in the case of the "midaccurate" subset, some fields were filled in by querying language models connected to the Internet.

The rest of "accurate" subset have been added by users at letrascarnavalcadiz.com.

Personal and Sensitive Information

The only sensitive information in the dataset is the names and surnames of the authors of the lyrics.

Considerations for Using the Data

Social Impact of Dataset

The use of this dataset has significant social impact.

Firstly, this dataset can positively contribute to the understanding and preservation of Cadiz's culture and traditions, as the Carnaval de Cádiz is an integral part of the city's cultural identity. By providing an accessible and easily searchable resource for carnival song lyrics, this dataset can assist cultural researchers, linguists, and the general public in better understanding and appreciating the rich tradition of the Carnaval de Cádiz.

Additionally, this dataset can be utilized to enhance natural language processing (NLP) technologies in Spanish, a language that can sometimes be underrepresented in NLP research. By providing a high-quality, culture-specific Spanish text corpus, this dataset can aid in improving the accuracy and cultural relevance of Spanish NLP models.

However, there are also risks associated with the use of this dataset. For instance, if used to train text generation models, these models could generate content that reinforces cultural stereotypes or perpetuates existing biases. Moreover, the automatic interpretation of carnival song lyrics can be challenging due to cultural and linguistic subtleties, and errors in this interpretation could lead to misunderstandings or misrepresentations of Cadiz's culture.

Finally, although this dataset does not contain a low-resource or underrepresented language, it does focus on a specific cultural tradition from a specific region of Spain. Therefore, its use can impact the Cadiz community by helping to preserve and disseminate its unique culture and traditions.

Discussion of Biases

The dataset is subject to several biases due to the nature of the data collection and the historical context of the Cadiz Carnival.

Firstly, there is a temporal bias in the dataset. More recent lyrics are overrepresented compared to older ones, as there is more information available on the internet about modern groups. This may lead to a skewed understanding of the evolution of the Carnival's themes over time.

Secondly, the dataset exhibits a popularity bias. Lyrics from more popular groups are overrepresented because individuals have chosen to write about them more frequently. This could potentially limit the diversity of styles and themes represented in the dataset.

Thirdly, there is a competition bias. Lyrics from groups that advanced further in the competition stages are overrepresented, resulting in more available lyrics from these groups. This might lead to an overemphasis on the styles and themes that tend to be more successful in the competition.

Lastly, the dataset reflects a gender bias. Given that there have historically been more male authors than female authors in the Cadiz Carnival, the majority of the dataset consists of lyrics written by men. This could potentially limit the representation of diverse perspectives and themes in the lyrics.

To mitigate these biases, we actively encourage the participation of the community. By verifying the different records of the dataset, reviewing the transcription of the lyrics and all the data for accuracy, and adding new lyrics, we hope to broaden the diversity and representation.

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

Iván Romero Reyna. Student of the Specialisation Course in Artificial Intelligence and Big Data at IES Rafael Alberti during the academic year 2022-2023.
Jesús Federico Franco Medinilla. Student of the Specialisation Course in Artificial Intelligence and Big Data at IES Rafael Alberti during the academic year 2022-2023.
Jesús Carlos Avecilla de la Herrán. Promoter in Cádiz AI.

Licensing Information

CC BY-SA 4.0

Citation Information

@misc{letrascarnavalcadiz2023,
  author = {Romero Reyna, Iván and Franco Medinilla, Jesús Federico and Avecilla de la Herrán, Jesús Carlos},
  title = {letras-carnaval-cadiz},
  year = {2023},
  url = {https://huggingface.co/datasets/IES-Rafael-Alberti/letras-carnaval-cadiz}
}

Contributions

Thanks to @ivanro, @jframed281 for adding this dataset.

Thanks to all the reviewers and contributors at letrascarnavalcadiz.com.