Unsupervised machine translation, which utilizes unpaired monolingual corpora as training data, has achieved comparable performance against supervised machine translation. However, it still suffers from data-scarce domains. To address this issue, this paper presents a novel meta-learning algorithm for unsupervised neural machine translation (UNMT) that trains the model to adapt to another domain by utilizing only a small amount of training data. We assume that domain-general knowledge is a significant factor in handling data-scarce domains. Hence, we extend the meta-learning algorithm, which utilizes knowledge learned from high-resource domains, to boost the performance of low-resource UNMT. Our model surpasses a transfer learning-based approach by up to 2-3 BLEU scores. Extensive experimental results show that our proposed algorithm is pertinent for fast adaptation and consistently outperforms other baselines.
Description
This paper introduces MetaUMT and MetaGUMT, novel meta-learning algorithms for unsupervised neural machine translation, focusing on low-resource domains. It demonstrates how these models can adapt to new domains with minimal training data, leveraging domain-general knowledge to handle data-scarce scenarios effectively.
%0 Journal Article
%1 Cheonbok2020
%A Park, Cheonbok
%A Tae, Yunwon
%A Kim, Taehee
%A Yang, Soyoung
%A Khan, Mohammad Azam
%A Park, Lucy
%A Choo, J.
%D 2020
%K computer-science edited_with_chatgpt low-resource-nlp machine-learning meta-learning neural-machine-translation posted_with_chatgpt unsupervised-learning
%P 2888-2901
%R 10.18653/v1/2021.acl-long.225
%T Unsupervised Neural Machine Translation for Low-Resource Domains via Meta-Learning
%U https://arxiv.org/abs/2010.09046
%X Unsupervised machine translation, which utilizes unpaired monolingual corpora as training data, has achieved comparable performance against supervised machine translation. However, it still suffers from data-scarce domains. To address this issue, this paper presents a novel meta-learning algorithm for unsupervised neural machine translation (UNMT) that trains the model to adapt to another domain by utilizing only a small amount of training data. We assume that domain-general knowledge is a significant factor in handling data-scarce domains. Hence, we extend the meta-learning algorithm, which utilizes knowledge learned from high-resource domains, to boost the performance of low-resource UNMT. Our model surpasses a transfer learning-based approach by up to 2-3 BLEU scores. Extensive experimental results show that our proposed algorithm is pertinent for fast adaptation and consistently outperforms other baselines.
@article{Cheonbok2020,
abstract = {Unsupervised machine translation, which utilizes unpaired monolingual corpora as training data, has achieved comparable performance against supervised machine translation. However, it still suffers from data-scarce domains. To address this issue, this paper presents a novel meta-learning algorithm for unsupervised neural machine translation (UNMT) that trains the model to adapt to another domain by utilizing only a small amount of training data. We assume that domain-general knowledge is a significant factor in handling data-scarce domains. Hence, we extend the meta-learning algorithm, which utilizes knowledge learned from high-resource domains, to boost the performance of low-resource UNMT. Our model surpasses a transfer learning-based approach by up to 2-3 BLEU scores. Extensive experimental results show that our proposed algorithm is pertinent for fast adaptation and consistently outperforms other baselines.},
added-at = {2023-11-22T11:29:47.000+0100},
author = {Park, Cheonbok and Tae, Yunwon and Kim, Taehee and Yang, Soyoung and Khan, Mohammad Azam and Park, Lucy and Choo, J.},
biburl = {https://www.bibsonomy.org/bibtex/2ebdf9c97e0861adf894821b67ba39a3c/tomvoelker},
day = 18,
description = {This paper introduces MetaUMT and MetaGUMT, novel meta-learning algorithms for unsupervised neural machine translation, focusing on low-resource domains. It demonstrates how these models can adapt to new domains with minimal training data, leveraging domain-general knowledge to handle data-scarce scenarios effectively.},
doi = {10.18653/v1/2021.acl-long.225},
interhash = {bcda68c44001fce64422926b78de3b9c},
intrahash = {ebdf9c97e0861adf894821b67ba39a3c},
keywords = {computer-science edited_with_chatgpt low-resource-nlp machine-learning meta-learning neural-machine-translation posted_with_chatgpt unsupervised-learning},
month = {10},
pages = {2888-2901},
timestamp = {2023-11-22T11:33:19.000+0100},
title = {Unsupervised Neural Machine Translation for Low-Resource Domains via Meta-Learning},
url = {https://arxiv.org/abs/2010.09046},
year = 2020
}