Spanish-BFF

Avaliable in Hugging Face https://huggingface.co/datasets/MMG/SpanishBFF

Dataset Description

Spanish-BFF is the first Spanish AI-generated dictionary using GPT3.

• Paper: Spanish Built Factual Freectianary (Spanish-BFF): the first AI-generated free dictionary
• Point of Contact: [email protected] , [email protected]

Dataset Summary

Spanish-BFF contains a total of 66353 lemmas with its definitions (only one definiton per lemma).

These lemmas correspond to nominal, adjetival, verbal and adverbial classes.

Languages

• Spanish (es)

Dataset Structure

Data Instances

{
'id': 'b0o8', 
'lemma': 'fomo', 
'definition': 'FOMO es un acrónimo de "miedo a perderse", y se refiere a la ansiedad que uno puede sentir cuando ve que otros están disfrutando de algo que él o ella no está haciendo.', 
}

Data Fields

{
  id: str
  lemma: str
  difinition: str
}

Data Splits

Split	Size
train	66,353

Content analysis

Number of nouns, adjetives, adverbs and verbs

• Number of nouns: 38093 (57.41 %)
• Number of adjetives: 17424 (26.26 %)
• Number of verbs: 9296 (14.01 %)
• Number of adverbs: 1540 (2.32 %)

Statistics

Uncertainties provided by a coverage factor k=1 within the standard deviation of the population of definitions.

• Total words in definitions: 551878
• Average words/definition: 8.3 +/- 5.1 words
• Average characters/definitions: 49.1 +/- 28.4 characters

Dataset Creation

Prompting

Each one of the definitons were generated in batches using the following prompt:

Generate in Spanish a definition of the word "[word]"

Considerations for Using the Data

Social Impact of Dataset

This corpus is the first open-source complete dictionary produced by LLMs. We intend to contribute to a better understanding and development of NLP and promote responsible use.

Biases and Hallucinations

This version has not been postprocessed to mitigate potential errors, biases or hallucinations the AI model could have generated.

Citation

@misc{https://doi.org/10.48550/arxiv.2302.12746,
  doi = {10.48550/ARXIV.2302.12746},
  
  url = {https://arxiv.org/abs/2302.12746},
  
  author = {Ortega-Martín, Miguel and García-Sierra, Óscar and Ardoiz, Alfonso and Armenteros, Juan Carlos and Álvarez, Jorge and Alonso, Adrián},
  
  keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
  
  title = {Spanish Built Factual Freectianary (Spanish-BFF): the first AI-generated free dictionary},
  
  publisher = {arXiv},
  
  year = {2023},
  
  copyright = {Creative Commons Attribution 4.0 International}
}