Pith. Sign in

RECORD

Exploring Cross-lingual Textual Style Transfer with Large Multilingual Language Models

PaperPDF from arXiv

data files

Paper breakdownThe authors' claims, conditions and measurements, as JSON Signed receiptProof this review has not been altered
X Bluesky LinkedIn Reddit Hacker News
Daniil Moskovskiy, Daryna Dementieva, Alexander Panchenko

Computation and Language · Submitted 5 Jun 2022

abstract compiled · 2026-09-23 · no language-model review

The authors' main claim However, models are not able to perform cross-lingual detoxification and direct fine-tuning on exact language is inevitable.

how it is stated · asserted 1.00 read the record →

Comment (0) Read on arXiv ↗
Cite ▾
arXiv ID2206.02252 (version 1)
Pith ID52ZXVUOE, Pith's permanent record of this paper and its review
CitationAlexander Panchenko, Daniil Moskovskiy, Daryna Dementieva. "Exploring Cross-lingual Textual Style Transfer with Large Multilingual Language Models." arXiv:2206.02252 (2022). https://pith.science/paper/52ZXVUOE
Claim paper

The compiled record

The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.

Open the compiled record

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-09-23: 1 request, 4,809 input tokens.

What the abstract itself states, as probabilities: claims first 0.18 · states limitation 0.93 · human participants 0.05 · controlled comparison 0.21 · randomized assignment 0.01 · states quantitative result 0.04 · compares with prior results 0.83 · mentions code or data availability 0.03.

The abstract, sentence by sentence

arXiv listing abstract v1. Results and the main claim are underlined; .

Detoxification is a task of generating text in polite style while preserving meaning and fluency of the original toxic text. Existing detoxification methods are designed to work in one exact language. This work investigates multilingual and cross-lingual detoxification and the behavior of large multilingual models like in this setting. Unlike previous works we aim to make large language models able to perform detoxification without direct fine-tuning in given language. Experiments show that multilingual models are capable of performing multilingual style transfer. However, models are not able to perform cross-lingual detoxification and direct fine-tuning on exact language is inevitable.

background contribution method result interpretation main claim

What the classifier reports for each sentence. The number is the probability it gave the label.
#sentencerolethe authors' own workstates the main conclusionstrength
s0 Detoxification is a task of generating text in polite style while preserving meaning and fluency of the original toxic text. background 1.00 0.05 0.02 asserted 1.00
s1 Existing detoxification methods are designed to work in one exact language. gap 0.60 0.04 0.03 asserted 1.00
s2 This work investigates multilingual and cross-lingual detoxification and the behavior of large multilingual models like in this setting. contribution 0.98 0.48 0.04 asserted 0.51
s3 Unlike previous works we aim to make large language models able to perform detoxification without direct fine-tuning in given language. contribution 0.98 0.25 0.07 asserted 0.83
s4 Experiments show that multilingual models are capable of performing multilingual style transfer. result 1.00 0.89 0.17 asserted 0.98
s5 However, models are not able to perform cross-lingual detoxification and direct fine-tuning on exact language is inevitable. result 0.68 0.89 0.76 asserted 1.00

What kind of paper the classifier reports this is: empirical study 1.00 · evidence from observation or experiment 1.00.

What was examined, and what was not

sources read
  • arXiv listing abstract v1
not examined
  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers
levels run
  • P abstract
not run
  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. Every number in the measurements table is the authors', checked by code. Record as JSON

The paper
Original abstract

Detoxification is a task of generating text in polite style while preserving meaning and fluency of the original toxic text. Existing detoxification methods are designed to work in one exact language. This work investigates multilingual and cross-lingual detoxification and the behavior of large multilingual models like in this setting. Unlike previous works we aim to make large language models able to perform detoxification without direct fine-tuning in given language. Experiments show that multilingual models are capable of performing multilingual style transfer. However, models are not able to perform cross-lingual detoxification and direct fine-tuning on exact language is inevitable.

Human review

How Pith checks papers → · Full audit of this paper →

How the literature uses it

1 arXiv paper cite this in 1 sentence

Each citing sentence is taken from the citing paper's LaTeX source and typed by Pith's classifier. A label appears at probability 0.8 or more (0.95 for support), a contest at 0.7 or more; the quoted words are the citing authors'. How each label was checked · JSON

What it cites · 3 sentences without a confident label

Uses the method or tool of 3 papers
  • Russian Content preservation score (SIM) is evaluated as a cosine similarity of LaBSE [ref] sentence embeddings.

    Language-agnostic BERT Sentence Embedding · Content Preservation · 1.00

  • English We measure fluency (FL) as a percentage of fluent sentences evaluated by the RoBERTa-based[<https://huggingface.co/roberta-large>] [ref] classifier of linguistic acceptability trained on CoLA [ref] dataset.

    RoBERTa: A Robustly Optimized BERT Pretraining Approach · Grammatic and language quality (fluency) · 0.99

  • We use Adam [ref] optimizer for fine-tuning with different learning rates ranging from $1 ·10^-3$ to $5 ·10^-5$ with linear learning rate scheduling.

    Adam: A Method for Stochastic Optimization · Training · 1.00

Cites as background 1 paper

Discussion

Know something about this paper? Leave a comment, or highlight any passage above to comment on it.