Spanish Pre-trained BERT Model and Evaluation Data

TL;DR

This paper introduces a Spanish-specific BERT model and a comprehensive set of evaluation tasks, improving performance on Spanish NLP benchmarks and providing resources for future research.

Contribution

The paper presents a new Spanish BERT model and a unified Spanish benchmark suite, facilitating better NLP performance and resource sharing for Spanish language processing.

Findings

The Spanish BERT model outperforms multilingual models on most tasks.

Achieved state-of-the-art results on several Spanish NLP benchmarks.

Public release of the model, data, and benchmark suite.

Abstract

The Spanish language is one of the top 5 spoken languages in the world. Nevertheless, finding resources to train or evaluate Spanish language models is not an easy task. In this paper we help bridge this gap by presenting a BERT-based language model pre-trained exclusively on Spanish data. As a second contribution, we also compiled several tasks specifically for the Spanish language in a single repository much in the spirit of the GLUE benchmark. By fine-tuning our pre-trained Spanish model, we obtain better results compared to other BERT-based models pre-trained on multilingual corpora for most of the tasks, even achieving a new state-of-the-art on some of them. We have publicly released our model, the pre-training data, and the compilation of the Spanish benchmarks.