| G. Vonitsanos, A. Kanavos, Ph. Mylonas |
| A Comparative Evaluation of NLP Frameworks for Spanish Part-of-Speech Tagging Using the CoNLL-2002 Corpus |
| IEEE International Conference on Big Data (IEEE BigData 2025), December 8-11, 2025, Macau, China |
|
ABSTRACT
|
| This paper presents a comparative study of three widely adopted Natural Language Processing (NLP) frameworks—NLTK, spaCy, and Stanza—for Part-of-Speech (POS) tagging in Spanish. Using the CoNLL-2002 corpus as a benchmark, the study evaluates each tool¢s performance in terms of accuracy, F1-score, execution time, and error distribution. The experimental design ensures consistency across datasets, preprocessing, and evaluation metrics, including unified XPOS→UPOS mapping for cross-tool comparability. Results reveal that NLTK¢s n-gram backoff tagger achieves the highest overall accuracy (97.16%) and computational efficiency, outperforming both neural models in speed and stability. Stanza demonstrates competitive F1 performance (87.4%) but incurs greater computational cost due to its BiLSTM architecture, while spaCy shows lower accuracy yet excels in processing speed, making it suitable for real-time applications. The findings highlight key trade-offs among statistical and neural approaches, emphasizing the influence of corpus alignment, model complexity, and linguistic adaptation. Future work will explore transformer-based models, multilingual evaluation, and explainable-AI integration to enhance interpretability and resource efficiency in POS tagging.
|
| 08 December , 2025 |
| G. Vonitsanos, A. Kanavos, Ph. Mylonas, "A Comparative Evaluation of NLP Frameworks for Spanish Part-of-Speech Tagging Using the CoNLL-2002 Corpus", IEEE International Conference on Big Data (IEEE BigData 2025), December 8-11, 2025, Macau, China |
[
BibTex] [
Print] [
 Back] |
