TransQuest: Translation Quality Estimation with Cross-lingual Transformers

The goal of quality estimation (QE) is to evaluate the quality of a translation without having access to a reference translation. High-accuracy QE that can be easily deployed for a number of language pairs is the missing piece in many commercial translation workflows as they have numerous potential uses. They can be employed to select the best translation when several translation engines are available or can inform the end user about the reliability of automatically translated content. In addition, QE systems can be used to decide whether a translation can be published as it is in a given context, or whether it requires human post-editing before publishing or translation from scratch by a human. The quality estimation can be done at different levels: document level, sentence level and word level.

With TransQuest, we have opensourced our research in translation quality estimation which also won the sentence-level direct assessment quality estimation shared task in WMT 2020 . TransQuest outperforms current open-source quality estimation frameworks such as OpenKiwi and DeepQuest .

Features

• Sentence-level translation quality estimation on both aspects: predicting post editing efforts and direct assessment.
• Word-level translation quality estimation capable of predicting quality of source words, target words and target gaps.
• Outperform current state-of-the-art quality estimation methods like DeepQuest and OpenKiwi in all the languages experimented.
• Pre-trained quality estimation models for fifteen language pairs are available in HuggingFace.

Installation

From pip

pip install transquest

From Source

git clone https://github.com/TharinduDR/TransQuest.git
cd TransQuest
pip install -r requirements.txt

Using Pre-trained Models

import torch
from transquest.algo.sentence_level.monotransquest.run_model import MonoTransQuestModel


model = MonoTransQuestModel("xlmroberta", "TransQuest/monotransquest-da-multilingual", num_labels=1, use_cuda=torch.cuda.is_available())
predictions, raw_outputs = model.predict([["Reducerea acestor conflicte este importantă pentru conservare.", "Reducing these conflicts is not important for preservation."]])
print(predictions)

Documentation

For more details follow the documentation.

Citations

If you are using the word-level architecture, please consider citing this paper which is accepted to ACL 2021 .

@InProceedings{ranasinghe2021,
author = {Ranasinghe, Tharindu and Orasan, Constantin and Mitkov, Ruslan},
title = {An Exploratory Analysis of Multilingual Word Level Quality Estimation with Cross-Lingual Transformers},
booktitle = {Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics},
year = {2021}
}

If you are using the sentence-level architectures, please consider citing these papers which were presented in COLING 2020 and in WMT 2020 at EMNLP 2020.

@InProceedings{transquest:2020a,
author = {Ranasinghe, Tharindu and Orasan, Constantin and Mitkov, Ruslan},
title = {TransQuest: Translation Quality Estimation with Cross-lingual Transformers},
booktitle = {Proceedings of the 28th International Conference on Computational Linguistics},
year = {2020}
}

@InProceedings{transquest:2020b,
author = {Ranasinghe, Tharindu and Orasan, Constantin and Mitkov, Ruslan},
title = {TransQuest at WMT2020: Sentence-Level Direct Assessment},
booktitle = {Proceedings of the Fifth Conference on Machine Translation},
year = {2020}
}