paraphrasing

This paper proposes a style-aware, prompt-driven approach for text anonymization using large language models to suppress identifiable stylistic markers while preserving meaning, reducing authorship attribution F1 by 60–70% and outperforming differential privacy-based methods.

The paper theoretically analyzes how repeated paraphrasing can align AI-generated text with human writing distributions, deriving convergence rates and implications for reliable AI-text detection.

Introduces ARB, a matched authorship-rewriting benchmark for evaluating AI-text detectors, showing that detector performance drops significantly when human text is rewritten by an LLM despite high recall on direct LLM-generated text.

This paper presents Dual-Embedding Watermarking (DEW), a semantic watermarking scheme for LLMs that improves robustness against paraphrasing and translation by leveraging contextual and token-level embeddings. Experimental results show improved detection after paraphrasing and translation compared to prior methods.

PASTA is a novel framework for knowledge updating in LLMs that combines data augmentation, question-answering generation, and self-learning DPO to integrate factual information from news articles, achieving accuracy improvement from 0.02 to 0.82 while preserving general capabilities.

This paper identifies that standard multiple-choice QA benchmarks are sensitive to phrasing artifacts, conflating knowledge with surface-form familiarity. The authors propose ParaEval, a framework that uses multiple paraphrases per answer option to score models based on most favorable phrasing, reducing false performance gaps and enabling more robust evaluation.

This paper investigates evasion attacks on machine-text detectors, finding that while current attacks degrade detector performance, stylistic fingerprints persist. A novel paraphrasing approach that mimics human styles can evade even style-based detectors, but multi-document analysis recovers detectability.

This paper presents an empirical study adapting the small language model Phi Silica for short-form text rewriting through dataset curation, prompt distillation, and parameter-efficient fine-tuning, showing that targeted adaptation significantly improves semantic fidelity and reduces hallucinations.

This paper reveals that commercial AI detectors like GPTZero and Pangram judge text from base language models as overwhelmingly human, while instruction-tuned model outputs are flagged as AI-generated. The authors propose HIP, a detector-agnostic iterative paraphrasing pipeline that improves human-likeness while preserving semantics.

A research paper finds that base language models appear human to AI detectors, unlike instruction-tuned models. The authors propose a paraphrasing pipeline (HIP) that improves human-likeness while preserving semantics across model sizes.