OPUS-MT – Building open translation services for the World

By Jörg Tiedemann, Santhosh Thottingal. 2020

737 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Towards Continual Knowledge Learning of Language Models (2021, methodology): “(3) The decontextualized sentence is paraphrased by a back-translation model (en→de→en) (Tiedemann & Thottingal, 2020) and checked whether the selected word is still in the paraphrased sentence; the sentence is discarded if not.”
Investigating Failures of Automatic Translation in the Case of Unambiguous Gender (2021, methodology): “1 Models We evaluate gendered translation of three pretrained open-source models, (i) OPUS-MT is a collection of 1000+ bilingual and multilingual (for certain translation directions) models (Tiedemann and Thottingal, 2020).”
LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion (2023, methodology): “For the Chinese-English translation task on WMT18 dataset, we use the public pre-trained opus-mt checkpoint (Tiedemann and Thottingal, 2020b).”
"Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models (2023, unknown): “In this experiment, we rely on Opus-MT [79] to convert jailbreak prompts from English to Chinese and back to English.”
MCoNaLa: A Benchmark for Code Generation from Multiple Natural Languages (2022, methodology): “To optimize translation quality, we compare three best performing MMT models: OPUSMT (Tiedemann and Thottingal, 2020), M2M-”
NL-Augmenter: A Framework for Task-Sensitive Natural Language Augmentation (2021, methodology): “This transformation is a generalization of the Back-Translation transformation to any sequence of languages supported by the Helsinki-NLP OpusMT models (Tiedemann and Thottingal, 2020).”
FrugalScore: Learning Cheaper, Lighter and Faster Evaluation Metrics for Automatic Text Generation (2021, methodology): “We used OPUS-MT because of its ready-to-use checkpoints available for many language pairs.”
One Question Answering Model for Many Languages with Cross-lingual Dense Passage Retrieval (2021, methodology): “Our translation models (to English and from English) are the pre-trained MarianMT (Junczys-Dowmunt et al., 2018) style OPUS-MT (Tiedemann and Thottingal, 2020) models available in Transformers library (Wolf et al., 2020) that are trained on the OPUS corpus (Tiedemann and Nygaard, 2004).”
English Intermediate-Task Training Improves Zero-Shot Cross-Lingual Transfer Too (2020, methodology): “To translate QQP and HellaSwag, we use pretrained machine translation models from OPUS-MT (Tiedemann and Thottingal, 2020).”
Sometimes We Want Ungrammatical Translations (2021, methodology): “3207 els —Helsinki/OPUS machine translation model (Tiedemann and Thottingal, 2020), the multilingual BART model (Liu et al.”
Collecting a Large-Scale Gender Bias Dataset for Coreference Resolution and Machine Translation (2021, background): “We used EasyNMT5 to evaluate three machine translation models: mBART50_m2m (Tang et al., 2020; Liu et al., 2020), m2m_100_418M (Fan et al., 2020), and Opus-MT (Tiedemann and Thottingal, 2020), representing the state-of-the-art for publicly available neural machine translations models.”
On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspective (2023, methodology): “We further adopt two ﬁne-tuned machine translation models from the Huggingface model hub: OPUS-MT-EN-ZH [48] and Trans-OPUS-MT-EN-ZH 5 .”
Samanantar: The Largest Publicly Available Parallel Corpora Collection for 11 Indic Languages (2021, background): “Some sentence aligned corpora were collected from OPUS3 (Tiedemann, 2012) on 21 March 2021.”
Asking Crowdworkers to Write Entailment Examples: The Best of Bad Options (2020, methodology): “We then translate the aligned non-English sentence into an English sentence, sÊ using the OPUS-MT (Tiedemann and Thottingal, 2020) machine translation systems, and treat (sE , sÊ) as a sentence pair.”
Pre-trained Summarization Distillation (2020, background): “(Liu et al., 2020b), “Marian” is trained from scratch on bilingual data (Tiedemann and Thottingal, 2020).”