RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation

Gabriele Sarti; Phu Mon Htut; Xing Niu; Benjamin Hsu; Anna Currey; Georgiana Dinu; Maria Nadejde

doi:10.18653/v1/2023.acl-short.126

RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation

Gabriele Sarti, Phu Mon Htut, Xing Niu, Benjamin Hsu, Anna Currey, Georgiana Dinu, Maria Nadejde

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

7 Citations (Scopus)

84 Downloads (Pure)

Abstract

Attribute-controlled translation (ACT) is a subtask of machine translation that involves controlling stylistic or linguistic attributes (like formality and gender) of translation outputs. While ACT has garnered attention in recent years due to its usefulness in real-world applications, progress in the task is currently limited by dataset availability, since most prior approaches rely on supervised methods. To address this limitation, we propose Retrieval and Attribute-Marking enhanced Prompting (RAMP), which leverages large multilingual language models to perform ACT in few-shot and zero-shot settings. RAMP improves generation accuracy over the standard prompting approach by (1) incorporating a semantic similarity retrieval component for selecting similar in-context examples, and (2) marking in-context examples with attribute annotations. Our comprehensive experiments show that RAMP is a viable approach in both zero-shot and few-shot settings.

Original language	English
Title of host publication	Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics
Editors	Anna Rogers, Jordan Boyd-Graber, Naoaki Okazaki
Place of Publication	Toronto, ON, CA
Publisher	Association for Computational Linguistics (ACL)
Pages	1476-1490
Number of pages	14
Volume	2
DOIs	https://doi.org/10.18653/v1/2023.acl-short.126
Publication status	Published - Jul-2023
Event	61st Annual Meeting of the Association for Computational Linguistics (ACL 2023) - 1 Harbour Square, Toronto, ON M5J 1A6, Canada, Toronto, Canada Duration: 9-Jul-2023 → 12-Jul-2023 https://2023.aclweb.org/

Conference

Conference	61st Annual Meeting of the Association for Computational Linguistics (ACL 2023)
Abbreviated title	ACL 2023
Country/Territory	Canada
City	Toronto
Period	09/07/2023 → 12/07/2023
Internet address	https://2023.aclweb.org/

Keywords

large language models
machine translation
Style transfer
natural language generation
prompting
information retrieval

Access to Document

10.18653/v1/2023.acl-short.126Licence: CC BY

RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled TranslationFinal publisher's version, 426 KBLicence: CC BY

Handle.net

Persistent link

Amazon AWS AI
Sarti, G. (Visiting researcher)
27-Jun-2022 → 30-Sept-2022
Activity: Visiting an external institution › Professional

Cite this

Sarti, G., Htut, P. M., Niu, X., Hsu, B., Currey, A., Dinu, G., & Nadejde, M. (2023). RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation. In A. Rogers, J. Boyd-Graber, & N. Okazaki (Eds.), Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Vol. 2, pp. 1476-1490). Association for Computational Linguistics (ACL). https://doi.org/10.18653/v1/2023.acl-short.126

Sarti, Gabriele ; Htut, Phu Mon ; Niu, Xing et al. / RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. editor / Anna Rogers ; Jordan Boyd-Graber ; Naoaki Okazaki. Vol. 2 Toronto, ON, CA : Association for Computational Linguistics (ACL), 2023. pp. 1476-1490

@inproceedings{b7f7c6d5f763406c88d2a346ac4c5555,

title = "RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation",

abstract = "Attribute-controlled translation (ACT) is a subtask of machine translation that involves controlling stylistic or linguistic attributes (like formality and gender) of translation outputs. While ACT has garnered attention in recent years due to its usefulness in real-world applications, progress in the task is currently limited by dataset availability, since most prior approaches rely on supervised methods. To address this limitation, we propose Retrieval and Attribute-Marking enhanced Prompting (RAMP), which leverages large multilingual language models to perform ACT in few-shot and zero-shot settings. RAMP improves generation accuracy over the standard prompting approach by (1) incorporating a semantic similarity retrieval component for selecting similar in-context examples, and (2) marking in-context examples with attribute annotations. Our comprehensive experiments show that RAMP is a viable approach in both zero-shot and few-shot settings.",

keywords = "large language models, machine translation, Style transfer, natural language generation, prompting, information retrieval",

author = "Gabriele Sarti and Htut, {Phu Mon} and Xing Niu and Benjamin Hsu and Anna Currey and Georgiana Dinu and Maria Nadejde",

year = "2023",

month = jul,

doi = "10.18653/v1/2023.acl-short.126",

language = "English",

volume = "2",

pages = "1476--1490",

editor = "Rogers, {Anna } and Jordan Boyd-Graber and Naoaki Okazaki",

booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics",

publisher = "Association for Computational Linguistics (ACL)",

note = "61st Annual Meeting of the Association for Computational Linguistics (ACL 2023), ACL 2023 ; Conference date: 09-07-2023 Through 12-07-2023",

url = "https://2023.aclweb.org/",

}

Sarti, G, Htut, PM, Niu, X, Hsu, B, Currey, A, Dinu, G & Nadejde, M 2023, RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation. in A Rogers, J Boyd-Graber & N Okazaki (eds), Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. vol. 2, Association for Computational Linguistics (ACL), Toronto, ON, CA, pp. 1476-1490, 61st Annual Meeting of the Association for Computational Linguistics (ACL 2023), Toronto, Ontario, Canada, 09/07/2023. https://doi.org/10.18653/v1/2023.acl-short.126

RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation. / Sarti, Gabriele; Htut, Phu Mon; Niu, Xing et al.
Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. ed. / Anna Rogers; Jordan Boyd-Graber; Naoaki Okazaki. Vol. 2 Toronto, ON, CA: Association for Computational Linguistics (ACL), 2023. p. 1476-1490.

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

TY - GEN

T1 - RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation

AU - Sarti, Gabriele

AU - Htut, Phu Mon

AU - Niu, Xing

AU - Hsu, Benjamin

AU - Currey, Anna

AU - Dinu, Georgiana

AU - Nadejde, Maria

PY - 2023/7

Y1 - 2023/7

N2 - Attribute-controlled translation (ACT) is a subtask of machine translation that involves controlling stylistic or linguistic attributes (like formality and gender) of translation outputs. While ACT has garnered attention in recent years due to its usefulness in real-world applications, progress in the task is currently limited by dataset availability, since most prior approaches rely on supervised methods. To address this limitation, we propose Retrieval and Attribute-Marking enhanced Prompting (RAMP), which leverages large multilingual language models to perform ACT in few-shot and zero-shot settings. RAMP improves generation accuracy over the standard prompting approach by (1) incorporating a semantic similarity retrieval component for selecting similar in-context examples, and (2) marking in-context examples with attribute annotations. Our comprehensive experiments show that RAMP is a viable approach in both zero-shot and few-shot settings.

AB - Attribute-controlled translation (ACT) is a subtask of machine translation that involves controlling stylistic or linguistic attributes (like formality and gender) of translation outputs. While ACT has garnered attention in recent years due to its usefulness in real-world applications, progress in the task is currently limited by dataset availability, since most prior approaches rely on supervised methods. To address this limitation, we propose Retrieval and Attribute-Marking enhanced Prompting (RAMP), which leverages large multilingual language models to perform ACT in few-shot and zero-shot settings. RAMP improves generation accuracy over the standard prompting approach by (1) incorporating a semantic similarity retrieval component for selecting similar in-context examples, and (2) marking in-context examples with attribute annotations. Our comprehensive experiments show that RAMP is a viable approach in both zero-shot and few-shot settings.

KW - large language models

KW - machine translation

KW - Style transfer

KW - natural language generation

KW - prompting

KW - information retrieval

U2 - 10.18653/v1/2023.acl-short.126

DO - 10.18653/v1/2023.acl-short.126

M3 - Conference contribution

VL - 2

SP - 1476

EP - 1490

BT - Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics

A2 - Rogers, Anna

A2 - Boyd-Graber, Jordan

A2 - Okazaki, Naoaki

PB - Association for Computational Linguistics (ACL)

CY - Toronto, ON, CA

T2 - 61st Annual Meeting of the Association for Computational Linguistics (ACL 2023)

Y2 - 9 July 2023 through 12 July 2023

ER -

Sarti G, Htut PM, Niu X, Hsu B, Currey A, Dinu G et al. RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation. In Rogers A, Boyd-Graber J, Okazaki N, editors, Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. Vol. 2. Toronto, ON, CA: Association for Computational Linguistics (ACL). 2023. p. 1476-1490 doi: 10.18653/v1/2023.acl-short.126

RAMP: Retrieval and Attribute-Marking Enhanced Prompting for Attribute-Controlled Translation

Abstract

Conference

Keywords

Access to Document

Handle.net

Fingerprint

Activities

Amazon AWS AI

Cite this