# mStyleDistance: Multilingual Style Embeddings and their Evaluation

Source: [mStyleDistance: Multilingual Style Embeddings and their Evaluation](<https://www.sota2.com/research/paper/68fa770c1e209131cea70be3>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T04:20:34.120Z

Authors: Justin Qiu, Jiacheng Zhu, Ajay Patel, Marianna Apidianaki, Chris Callison-Burch

Published: 2025-02-21T03:11:41.000Z

## Abstract

Style embeddings are useful for stylistic analysis and style transfer; however, only English style embeddings have been made available. We introduce Multilingual StyleDistance (mStyleDistance), a multilingual style embedding model trained using synthetic data and contrastive learning. We train the model on data from nine languages and create a multilingual STEL-or-Content benchmark (Wegmann et al., 2022) that serves to assess the embeddings' quality. We also employ our embeddings in an authorship verification task involving different languages. Our results show that mStyleDistance embeddings outperform existing models on these multilingual style benchmarks and generalize well to unseen features and languages. We make our model publicly available at https://huggingface.co/StyleDistance/mstyledistance .

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Style Representation Evaluation on STEL-or-Content Multilingual (averaged across languages)](<https://www.sota2.com/research/sota/style-representation-evaluation-on-stel-or-content-multilingual-averaged-across-languages.md>): Style Representation Evaluation; STEL-or-Content Multilingual (averaged across languages); Simplicity Score: 36.

- [Style Representation Evaluation on STEL-or-Content Cross-lingual (averaged across languages)](<https://www.sota2.com/research/sota/style-representation-evaluation-on-stel-or-content-cross-lingual-averaged-across-languages.md>): Style Representation Evaluation; STEL-or-Content Cross-lingual (averaged across languages); Formality: 0.53.

- [Authorship Verification on PAN AV 2015 (test)](<https://www.sota2.com/research/sota/authorship-verification-on-pan-av-2015-test.md>): Authorship Verification; PAN AV 2015 (test); ROC-AUC (Greek): 0.58.

- [Authorship Verification on PAN Average 2013-2015 (test)](<https://www.sota2.com/research/sota/authorship-verification-on-pan-average-2013-2015-test.md>): Authorship Verification; PAN Average 2013-2015 (test); Greek Avg ROC-AUC: 0.64.

- [Authorship Verification on PAN AV 2014 (test)](<https://www.sota2.com/research/sota/authorship-verification-on-pan-av-2014-test.md>): Authorship Verification; PAN AV 2014 (test); ROC-AUC (Greek): 0.53.

- [Authorship Verification on PAN AV 2013 (test)](<https://www.sota2.com/research/sota/authorship-verification-on-pan-av-2013-test.md>): Authorship Verification; PAN AV 2013 (test); ROC-AUC (Greek): 0.41.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2502.15168>)

[PDF](<https://arxiv.org/pdf/2502.15168.pdf>)

[Implementation](<https://huggingface.co/StyleDistance/mstyledistance>)