Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain

Miguel Angel Rios Gaona; Raluca-Maria Chereji; Alina Secara; Dragos Ioan Ciobanu

Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain

Miguel Angel Rios Gaona, Raluca-Maria Chereji, Alina Secara, Dragos Ioan Ciobanu

Department for Translation Studies

Publications: Contribution to book › Contribution to proceedings › Peer Reviewed

Abstract

Multilingual Neural Machine Translation (MNMT) models leverage many language pairs during training to improve translation quality for low-resource languages by transferring knowledge from high-resource languages. We study the quality of a domain-adapted MNMT model in the medical domain for English-Romanian with automatic metrics and a human error typology annotation which includes terminology-specific error categories. We compare the out-of-domain MNMT with the in-domain adapted MNMT. The in-domain MNMT model outperforms the out-of-domain MNMT in all measured automatic metrics and produces fewer terminology errors.

Original language	English
Title of host publication	Proceedings of the 44th Translating and the Computer (TC44) conference
Editors	Joss Moorkens, Vilelmini Sosoni
Place of Publication	Geneva
Publisher	Editions Tradulex
Pages	27-38
ISBN (Electronic)	978-2-9701733-0-4
Publication status	Published - Sep 2023
Event	Translating and the Computer - European Convention Center Luxembourg (ECCL), Luxembourg, Luxembourg Duration: 24 Nov 2022 → 25 Nov 2022 Conference number: 44 https://asling.org/tc44/

Conference

Conference	Translating and the Computer
Abbreviated title	TC44
Country/Territory	Luxembourg
City	Luxembourg
Period	24/11/22 → 25/11/22
Internet address	https://asling.org/tc44/

Austrian Fields of Science 2012

602051 Translation studies

Keywords

Neural Machine Translation
domain adaptation
automatic metrics
error annotation
medical domain
English-Romanian

Access to Document

https://www.tradulex.com/varia/TC44-luxembourg2022.pdf

Cite this

@inproceedings{d0a4fffd3dab45049ab72c2a937457c9,

title = "Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain",

abstract = "Multilingual Neural Machine Translation (MNMT) models leverage many language pairs during training to improve translation quality for low-resource languages by transferring knowledge from high-resource languages. We study the quality of a domain-adapted MNMT model in the medical domain for English-Romanian with automatic metrics and a human error typology annotation which includes terminology-specific error categories. We compare the out-of-domain MNMT with the in-domain adapted MNMT. The in-domain MNMT model outperforms the out-of-domain MNMT in all measured automatic metrics and produces fewer terminology errors.",

keywords = "Neural Machine Translation, domain adaptation, automatic metrics, error annotation, medical domain, English-Romanian",

author = "{Rios Gaona}, {Miguel Angel} and Raluca-Maria Chereji and Alina Secara and Ciobanu, {Dragos Ioan}",

year = "2023",

month = sep,

language = "English",

pages = "27--38",

editor = "Joss Moorkens and Vilelmini Sosoni",

booktitle = "Proceedings of the 44th Translating and the Computer (TC44) conference",

publisher = "Editions Tradulex",

note = "Translating and the Computer, TC44 ; Conference date: 24-11-2022 Through 25-11-2022",

url = "https://asling.org/tc44/",

}

Rios Gaona, MA , Chereji, R-M , Secara, A & Ciobanu, DI 2023, Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain. in J Moorkens & V Sosoni (eds), Proceedings of the 44th Translating and the Computer (TC44) conference. Editions Tradulex, Geneva, pp. 27-38, Translating and the Computer, Luxembourg, Luxembourg, 24/11/22. <https://www.tradulex.com/varia/TC44-luxembourg2022.pdf>

Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain. / Rios Gaona, Miguel Angel ; Chereji, Raluca-Maria ; Secara, Alina et al.

Proceedings of the 44th Translating and the Computer (TC44) conference. ed. / Joss Moorkens; Vilelmini Sosoni. Geneva : Editions Tradulex, 2023. p. 27-38.

Publications: Contribution to book › Contribution to proceedings › Peer Reviewed

TY - GEN

T1 - Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain

AU - Rios Gaona, Miguel Angel

AU - Chereji, Raluca-Maria

AU - Secara, Alina

AU - Ciobanu, Dragos Ioan

N1 - Conference code: 44

PY - 2023/9

Y1 - 2023/9

N2 - Multilingual Neural Machine Translation (MNMT) models leverage many language pairs during training to improve translation quality for low-resource languages by transferring knowledge from high-resource languages. We study the quality of a domain-adapted MNMT model in the medical domain for English-Romanian with automatic metrics and a human error typology annotation which includes terminology-specific error categories. We compare the out-of-domain MNMT with the in-domain adapted MNMT. The in-domain MNMT model outperforms the out-of-domain MNMT in all measured automatic metrics and produces fewer terminology errors.

AB - Multilingual Neural Machine Translation (MNMT) models leverage many language pairs during training to improve translation quality for low-resource languages by transferring knowledge from high-resource languages. We study the quality of a domain-adapted MNMT model in the medical domain for English-Romanian with automatic metrics and a human error typology annotation which includes terminology-specific error categories. We compare the out-of-domain MNMT with the in-domain adapted MNMT. The in-domain MNMT model outperforms the out-of-domain MNMT in all measured automatic metrics and produces fewer terminology errors.

KW - Neural Machine Translation

KW - domain adaptation

KW - automatic metrics

KW - error annotation

KW - medical domain

KW - English-Romanian

M3 - Contribution to proceedings

SP - 27

EP - 38

BT - Proceedings of the 44th Translating and the Computer (TC44) conference

A2 - Moorkens, Joss

A2 - Sosoni, Vilelmini

PB - Editions Tradulex

CY - Geneva

T2 - Translating and the Computer

Y2 - 24 November 2022 through 25 November 2022

ER -

Impact of Domain-Adapted Multilingual Neural Machine Translation in the Medical Domain

Abstract

Conference

Austrian Fields of Science 2012

Keywords

Access to Document

Fingerprint

Cite this