ParaGuide: Guided Diffusion Paraphrasers for Plug-and-Play Textual Style Transfer

Authors: Zachary Horvitz, Ajay Patel, Chris Callison-Burch, Zhou Yu, Kathleen McKeown

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate the method on the Enron Email Corpus, with both human and automatic evaluations, and find that it outperforms strong baselines on formality, sentiment, and even authorship style transfer.
Researcher Affiliation Academia 1 Columbia University 2 University of Pennsylvania
Pseudocode Yes Algorithm 1: Para Guide Style Transfer
Open Source Code Yes Our code is publicly available at https://github.com/ zacharyhorvitz/Para Guide.
Open Datasets Yes We evaluate our method on the Enron Email Corpus, which comprises several hundred thousand emails made public during the US government s investigation of Enron (Klimt and Yang 2004; Peterson, Hohensee, and Xia 2011). ... In addition to the Enron corpus, we also build a pretraining corpus from the Reddit Million User Dataset (MUD) (Andrews and Bishop 2019; Khan et al. 2021)
Dataset Splits Yes To build our training and validation datasets for attribute style transfer, we use popular existing formality and sentiment classifiers to score texts from the holdout authors in the Enron dataset.
Hardware Specification No The paper states that the method "can be fine-tuned on a single GPU" but does not specify the model or type of GPU, or any other hardware components.
Software Dependencies No The paper mentions software like SSD-LM RoBERTa-Large, but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes PARAGUIDE s complete inference procedure is specified in Algorithm 1. ... We additionally visualize the affect of varying λ on authorship style transfer in Figure 2. When λ is small, the paraphrase-conditioned diffusion model reconstructs a more semantically faithful, fluent output. However, we can increase λ to improve Confusion scores, at the cost of semantic consistency and fluency. At the lowest setting, PARAGUIDE s Fluency and Similarity score are similar to those of Chat GPT-3.5 (0.78 vs 0.79 and 0.52 vs 0.56).