Data Selection for Unsupervised Translation of German--Upper Sorbian

Lukas Edman; Antonio Toral; Gertjan van Noord

Data Selection for Unsupervised Translation of German--Upper Sorbian

Lukas Edman, Antonio Toral, Gertjan van Noord

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

63 Downloads (Pure)

Abstract

This paper describes the methods behind the systems submitted by the University of Groningen for the WMT 2020 Unsupervised Machine Translation task for German--Upper Sorbian. We investigate the usefulness of data selection in the unsupervised setting. We find that we can perform data selection using a pretrained model and show that the quality of a set of sentences or documents can have a great impact on the performance of the UNMT system trained on it. Furthermore, we show that document-level data selection should be preferred for training the XLM model when possible. Finally, we show that there is a trade-off between quality and quantity of the data used to train UNMT systems.

Original language	English
Title of host publication	Proceedings of the Fifth Conference on Machine Translation (WMT)
Publisher	Association for Computational Linguistics, ACL Anthology
Pages	1099-1103
Number of pages	5
Publication status	Published - Nov-2020
Event	Fifth Conference on Machine Translation - Online Duration: 19-Nov-2020 → 20-Nov-2020

Conference

Conference	Fifth Conference on Machine Translation
Abbreviated title	WMT20
Period	19/11/2020 → 20/11/2020

Access to Document

Data Selection for Unsupervised Translation of German–Upper SorbianFinal publisher's version, 216 KBLicence: CC BY

Handle.net

Persistent link

Cite this

@inproceedings{a8807965a6d54ac590f7321995955ebc,

title = "Data Selection for Unsupervised Translation of German--Upper Sorbian",

abstract = "This paper describes the methods behind the systems submitted by the University of Groningen for the WMT 2020 Unsupervised Machine Translation task for German--Upper Sorbian. We investigate the usefulness of data selection in the unsupervised setting. We find that we can perform data selection using a pretrained model and show that the quality of a set of sentences or documents can have a great impact on the performance of the UNMT system trained on it. Furthermore, we show that document-level data selection should be preferred for training the XLM model when possible. Finally, we show that there is a trade-off between quality and quantity of the data used to train UNMT systems.",

author = "Lukas Edman and Antonio Toral and \{van Noord\}, Gertjan",

year = "2020",

month = nov,

language = "English",

pages = "1099--1103",

booktitle = "Proceedings of the Fifth Conference on Machine Translation (WMT)",

publisher = "Association for Computational Linguistics, ACL Anthology",

note = "Fifth Conference on Machine Translation, WMT20 ; Conference date: 19-11-2020 Through 20-11-2020",

}

Data Selection for Unsupervised Translation of German--Upper Sorbian. / Edman, Lukas; Toral, Antonio; van Noord, Gertjan.
Proceedings of the Fifth Conference on Machine Translation (WMT). Association for Computational Linguistics, ACL Anthology, 2020. p. 1099-1103.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Data Selection for Unsupervised Translation of German--Upper Sorbian

AU - Edman, Lukas

AU - Toral, Antonio

AU - van Noord, Gertjan

PY - 2020/11

Y1 - 2020/11

N2 - This paper describes the methods behind the systems submitted by the University of Groningen for the WMT 2020 Unsupervised Machine Translation task for German--Upper Sorbian. We investigate the usefulness of data selection in the unsupervised setting. We find that we can perform data selection using a pretrained model and show that the quality of a set of sentences or documents can have a great impact on the performance of the UNMT system trained on it. Furthermore, we show that document-level data selection should be preferred for training the XLM model when possible. Finally, we show that there is a trade-off between quality and quantity of the data used to train UNMT systems.

AB - This paper describes the methods behind the systems submitted by the University of Groningen for the WMT 2020 Unsupervised Machine Translation task for German--Upper Sorbian. We investigate the usefulness of data selection in the unsupervised setting. We find that we can perform data selection using a pretrained model and show that the quality of a set of sentences or documents can have a great impact on the performance of the UNMT system trained on it. Furthermore, we show that document-level data selection should be preferred for training the XLM model when possible. Finally, we show that there is a trade-off between quality and quantity of the data used to train UNMT systems.

M3 - Conference contribution

SP - 1099

EP - 1103

BT - Proceedings of the Fifth Conference on Machine Translation (WMT)

PB - Association for Computational Linguistics, ACL Anthology

T2 - Fifth Conference on Machine Translation

Y2 - 19 November 2020 through 20 November 2020

ER -

Data Selection for Unsupervised Translation of German--Upper Sorbian

Abstract

Conference

Access to Document

Handle.net

Fingerprint

Cite this