TaG-GEC: A Supervised Formality Style Transfer Framework via Tag and Generate with Grammar Error Correction

Author

• Paolo Edni V. Espiritu, De La Salle UniversityFollow

Date of Publication

2025

Document Type

Dissertation/Thesis

Degree Name

Master of Science in Computer Science

College

College of Computer Studies

Department/Unit

Software Technology

Thesis Advisor

Charibeth K. Cheng

Defense Panel Chair

Ethel Ong

Defense Panel Member

Shirley Chu
Charibeth Cheng

Abstract (English)

Formal style transfer (FST) for the Filipino language remains an evolving research area, with existing models struggling to preserve the semantic context of the input. Additionally, prior evaluations of Filipino FST models lack clarity due to the absence of standardized and comprehensive assessment metrics. This study addresses these challenges by (1) establishing a manually annotated test set as a benchmark for future FST research, (2) introducing a joint evaluation framework that incorporates formality, BERTScore, and perplexity for a more holistic assessment, and (3) proposing the TaG-GEC pipeline and its variations to enhance FST model performance. Through benchmarking on the newly established test set, this study identifies GEC-TaG v2 as the best-performing model, achieving the highest formality improvements while maintaining fluency. Results indicate a trade-off between formality and BERTScore, where higher formality scores often come at the expense of slight reductions in semantic similarity. These findings contribute to the development of more structured evaluations and effective Filipino FST models.

Abstract Format

html

Language

English

Recommended Citation

Espiritu, P. V. (2025). TaG-GEC: A Supervised Formality Style Transfer Framework via Tag and Generate with Grammar Error Correction. Retrieved from https://animorepository.dlsu.edu.ph/etdm_softtech/17

Upload Full Text

wf_yes

Embargo Period

4-22-2025