Learning Text Styles: A Study on Transfer, Attribution, and Verification

TL;DR

This thesis explores advanced methods for manipulating, attributing, and verifying text styles using large language models, focusing on style transfer, authorship attribution, and authorship verification with improved techniques.

Contribution

It introduces parameter-efficient adaptation, contrastive disentanglement, and instruction-based fine-tuning to enhance style transfer, attribution, and verification tasks.

Findings

Improved style transfer quality and content preservation.

Enhanced accuracy in authorship attribution and verification.

Effective use of large language models with new adaptation techniques.

Abstract

This thesis advances the computational understanding and manipulation of text styles through three interconnected pillars: (1) Text Style Transfer (TST), which alters stylistic properties (e.g., sentiment, formality) while preserving content; (2)Authorship Attribution (AA), identifying the author of a text via stylistic fingerprints; and (3) Authorship Verification (AV), determining whether two texts share the same authorship. We address critical challenges in these areas by leveraging parameter-efficient adaptation of large language models (LLMs), contrastive disentanglement of stylistic features, and instruction-based fine-tuning for explainable verification.