aiqbench analysis
This research introduces RT-SFT, a novel approach to text style transfer that leverages large-scale neural machine translation systems as implicit normalizers to generate pseudo-parallel training data. By LoRA-finetuning instruction-tuned LLMs on this synthesized dataset, the method achieves superior performance across multiple domains compared to standard few-shot in-context learning techniques. The study effectively repurposes existing massive multilingual models as a scalable tool for creating high-quality supervision signals without requiring expensive parallel corpora.
RT-SFT: Text Style Transfer from Non-Parallel Corpora by Roundtrip Translation
arXiv:2602.15013v2 Announce Type: replace Abstract: Text style transfer (TST) is naturally a supervised task - rewrite a sentence in a target style while preserving its meaning - yet the parallel corpora that supervision requires exist for only a handful of style domains. A common workaround is to *normalize* an input into a style-agnostic intermediate and then *stylize* it into the target style, but the normalizer is typically a lightweight, task-specific paraphraser applied only at test time, feeding a correspondingly small stylizer. We observe that a style-stripping normalizer already exists at scale: neural MT systems trained on hundreds of millions of general-domain sentence pairs preserve content while regressing toward generic phrasing, so roundtrip translation through a pivot language strips stylistic signal without any task-specific training. This turns normalization from an inference-time patch into a data-generation tool. Roundtrip-translating a monolingual in-style corpus yields a pseudo-parallel corpus on which we LoRA-finetune an instruction-tuned LLM as the stylizer (RT-SFT); applying the same normalizer to test queries keeps that stylizer in-distribution. We show that across four style domains, RT-SFT outperforms state-of-the-art methods, such as few-shot in-context learning, by considerable margins. We also report on effective retrieval augmentation methods for expert style domains with strict terminology and naming conventions.
Abstract:Text style transfer (TST) is naturally a supervised task - rewrite a sentence in a target style while preserving its meaning - yet the parallel corpora that supervision requires exist for only a handful of style domains. A common workaround is to *normalize* an input into a style-agnostic intermediate and then *stylize* it into the target style, but the normalizer is typically a lightweight, task-specific paraphraser applied only at test time, feeding a correspondingly small stylizer. We observe that a style-stripping normalizer already exists at scale: neural MT systems trained on hundreds of millions of general-domain sentence pairs preserve content while regressing toward generic phrasing, so roundtrip translation through a pivot language strips stylistic signal without any task-specific training. This turns normalization from an inference-time patch into a data-generation tool. Roundtrip-translating a monolingual in-style corpus yields a pseudo-parallel corpus on which we LoRA-finetune an instruction-tuned LLM as the stylizer (RT-SFT); applying the same normalizer to test queries keeps that stylizer in-distribution. We show that across four style domains, RT-SFT outperforms state-of-the-art methods, such as few-shot in-context learning, by considerable margins. We also report on effective retrieval augmentation methods for expert style domains with strict terminology and naming conventions.
Computation and Language (cs.CL)
arXiv:2602.15013 [cs.CL]
(or arXiv:2602.15013v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2602.15013 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 291 words
Related coverage
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
Evaluating AI Agents Live at the Grounded Reasoning Cup
Databricks · Aug 18, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026
Training Azerbaijani language models on Amazon SageMaker AI
AWS · May 28, 2026